diagnostics

Systematic Review
Artificial Intelligence-Based Applications for Bone Fracture
Detection Using Medical Images: A Systematic Review
Mohammed Kutbi

College of Computing and Informatics, Saudi Electronic University, Riyadh 13316, Saudi Arabia;
m.kutbi@seu.edu.sa

Abstract: Artificial intelligence (AI) is making notable advancements in the medical field, particularly
in bone fracture detection. This systematic review compiles and assesses existing research on AI
applications aimed at identifying bone fractures through medical imaging, encompassing studies
from 2010 to 2023. It evaluates the performance of various AI models, such as convolutional neural
networks (CNNs), in diagnosing bone fractures, highlighting their superior accuracy, sensitivity,
and specificity compared to traditional diagnostic methods. Furthermore, the review explores the
integration of advanced imaging techniques like 3D CT and MRI with AI algorithms, which has
led to enhanced diagnostic accuracy and improved patient outcomes. The potential of Generative
AI and Large Language Models (LLMs), such as OpenAI’s GPT, to enhance diagnostic processes
through synthetic data generation, comprehensive report creation, and clinical scenario simulation is
also discussed. The review underscores the transformative impact of AI on diagnostic workflows
and patient care, while also identifying research gaps and suggesting future research directions to
enhance data quality, model robustness, and ethical considerations.

Keywords: bone fracture; image classification; medical images

1. Introduction

Citation: Kutbi, M. Artificial

Medical records are maintained in diverse formats, including textual descriptions,
Intelligence-Based Applications for
audio recordings, and visual data. Text data includes details about diseases, symptoms,
Bone Fracture Detection Using
diagnoses, and treatments. Audio data primarily consists of recorded conversations be-
Medical Images: A Systematic Review. tween healthcare providers and patients. Visual data comprises medical images such as
Diagnostics 2024, 14, 1879. X-rays, CT scans, and MRIs, as well as videos of patients performing physical therapy
https://doi.org/10.3390/ exercises. The development of digital medical record systems like the Picture Archiving and
diagnostics14171879 Communication System (PACS) has greatly enhanced the accessibility and management of
these varied data types [1–3].
Academic Editor: Evangelos Terpos
Bone fractures are frequent injuries that necessitate swift diagnosis and treatment.
Received: 15 July 2024 Although imaging techniques such as X-rays and CT scans are effective for detecting
Revised: 19 August 2024 fractures, the manual interpretation of these images is often time-consuming, error-prone,
Accepted: 26 August 2024 and dependent on the clinician’s expertise and experience [4,5]. AI applications offer the
Published: 27 August 2024 potential to enhance the accuracy and efficiency of bone fracture detection by automating
parts of the diagnostic process [6–8]. Recently, there has been a significant increase in
research focused on various AI technologies for bone fracture identification [9,10].
Convolutional neural networks (CNNs) and other deep learning techniques have
Copyright: © 2024 by the author.
been effectively utilized across many fields, including healthcare [11–13]. Deep learning,
Licensee MDPI, Basel, Switzerland.
a branch of machine learning, excels in analyzing medical images by learning from large
This article is an open access article
datasets to perform specific tasks. For example, while doctors diagnose bone fractures by
distributed under the terms and
conditions of the Creative Commons
visually examining X-rays, deep learning models can be trained to perform these diagnostic
Attribution (CC BY) license (https://
tasks using extensive datasets of bone images [14–16].
creativecommons.org/licenses/by/
This systematic review aims to compile and evaluate current research on AI-based
4.0/). applications for bone fracture detection using medical images. By reviewing studies from

Diagnostics 2024, 14, 1879. https://doi.org/10.3390/diagnostics14171879 https://www.mdpi.com/journal/diagnostics

Diagnostics 2024, 14, 1879 2 of 20

the past decade, this review provides a comprehensive overview of the effectiveness of
various AI models, their applications in clinical settings, and the challenges that must be
addressed for successful implementation. Additionally, the review identifies gaps in the
existing research, and suggests potential directions for future studies. The methodology
follows the guidelines established by Okoli and Schabram, ensuring a thorough and
systematic literature review [17,18].
In accordance with Okoli and Schabram’s guidelines [17], our review process involved
a detailed examination of the literature on AI-based bone fracture detection. The objectives
include summarizing advanced applications, identifying challenges, and highlighting
potential areas for further research. The review process began by defining search databases
and keywords, followed by collecting relevant articles. Pre-defined inclusion and exclusion
criteria were then applied to the collected articles. Our findings are presented in this review,
with more details regarding our methodology available in Section 3.
Recently, there has been a significant increase in research focused on various AI
technologies for bone fracture identification. However, existing studies predominantly
emphasize traditional imaging techniques and established AI models, leaving critical
gaps in understanding the broader implications of AI integration with more advanced
imaging modalities like 3D CT and MRI. Moreover, while AI’s diagnostic accuracy has been
explored, there is less comprehensive analysis of the additional benefits AI offers, such
as improvements in diagnostic efficiency, cost-effectiveness, and accessibility, particularly
when these technologies are deployed in clinical settings.
This systematic review seeks to address these gaps by evaluating the effectiveness
of AI in bone fracture detection across various imaging modalities, identifying the key
benefits these technologies bring to clinical practice, and examining the specific applications
and challenges associated with their implementation. By also considering the potential
of emerging technologies such as Generative AI and Large Language Models (LLMs),
this review aims to provide a holistic understanding of how AI is transforming bone
fracture detection and what obstacles must be overcome to fully realize its potential in
clinical environments.

2. Related Work
Over the past decade, the application of Artificial Intelligence (AI) in medical imag-
ing has advanced significantly, with numerous studies highlighting the potential of AI,
particularly deep learning models, to improve diagnostic accuracy and efficiency.
Rajpurkar et al. [19] developed CheXNet, a deep convolutional neural network (CNN)
with 121 layers, which outperformed radiologists in detecting pneumonia from chest X-
rays. This model, trained on a dataset of over 100,000 chest X-ray images, exemplifies the
effectiveness of deep learning in medical imaging.
For liver disease diagnosis, Lin et al. [20] utilized Classification and Regression Trees
(CART) alongside Case-Based Reasoning (CBR) to create a diagnostic model. Their two-step
process first used CART to detect the presence of liver disease, followed by CBR to specify
the type of liver disease, demonstrating AI’s role in enhancing diagnostic precision and
supporting clinical decisions.
Dombi et al. [21] employed an artificial neural network (ANN) to predict the outcomes
of rib fractures using patient records. Their model evaluated 20 intake variables to predict
four outcome variables: hospital stay length, ICU days, survival, and mortality. The
ANN achieved up to 98% accuracy, showcasing AI’s potential in early prediction and
patient management.
In musculoskeletal imaging, Zhang et al. [22] introduced TandemNet, a framework
integrating linguistic understanding into medical image analysis. This model combines
textual and visual data to improve medical report analysis, enhancing the interpretability
and accuracy of deep learning models.
Ypsilantis and Montana [23] developed a recurrent neural network (RNN) that focuses
on relevant image areas to diagnose conditions like bone fractures. This model uses a
Diagnostics 2024, 14, 1879 3 of 20

recurrent visual attention mechanism to efficiently identify regions of interest, achieving

high diagnostic accuracy with fewer parameters.
Fu et al. [24] proposed a visualization technique for CT scans that maintains the 3D
proximal femur’s texture and structure while offering a comprehensive view of the fracture
line. This method aids in detecting fractures and assists radiologists in identifying complex
fracture patterns.
Yaqub et al. [25] presented an unsupervised machine learning approach for clustering
unlabeled fetal ultrasound images. Their method targets regions with significant anatomical
structures, achieving high categorization accuracy on a large dataset of clinical ultrasound
images, addressing challenges related to varying image quality and fetal positioning.
Recent work by Rajpurkar et al. [26] led to the development of extensive, high-quality
datasets, driving advancements in AI applications in medical imaging. The MURA dataset,
which includes 40,561 images from 14,863 upper extremity studies labeled by radiologists,
supports robust abnormality detection model development.
Systematic reviews have been valuable in consolidating research findings and iden-
tifying future research directions across various fields. Alammary et al. [27] conducted a
review on blended learning models for introductory programming courses, highlighting
effective strategies and research gaps. Liang and Ji [28] reviewed privacy challenges in
IoT-based blockchain systems, providing comprehensive insights and suggesting future
research areas. Konttila et al. [29] reviewed healthcare professionals’ digital competence,
while AlShamsi et al. [30] focused on blockchain adoption.
Several studies have made significant contributions to fracture detection using AI.
Meena and Roy [31] reviewed advances in bone fracture detection using deep supervised
learning, emphasizing improvements in diagnostic accuracy and efficiency. Kim and
MacKinnon [32] investigated transfer learning in fracture detection, demonstrating the
adaptation of pre-trained models for specific medical imaging tasks. Chung et al. [33]
developed a deep learning algorithm for detecting and classifying proximal humerus
fractures. Urakawa et al. [34] used a deep CNN to detect intertrochanteric hip fractures
with orthopedist-level accuracy. Yu et al. [35] demonstrated AI integration into routine
diagnostics for hip fracture detection.
Choi et al. [36] utilized a dual-input CNN for automated pediatric supracondylar
fracture detection. Majkowska et al. [37] assessed chest radiograph interpretation using
deep learning models, emphasizing the need for radiologist-adjudicated reference stan-
dards. Johari et al. [38] developed a probabilistic neural network for detecting vertical root
fractures in premolar teeth.
Heimer et al. [39] used deep learning to classify skull fractures on curved maxi-
mum intensity projections. Kitamura et al. [40] developed a CNN ensemble for ankle
fracture detection. Gan et al. [41] compared AI detection of distal radius fractures with
professional assessments.
Lindsey et al. [42] demonstrated deep neural networks’ ability to enhance clinician
fracture detection. Adams et al. [43] compared deep learning with perceptual training for
neck of femur fracture detection. Rahmaniar and Wang [44] created a real-time automated
system for segmenting and classifying calcaneal fractures in CT images.
Tomita et al. [45] focused on detecting osteoporotic vertebral fractures using deep
neural networks on CT scans. Muehlematter et al. [46] applied texture analysis and machine
learning to detect vertebrae at risk of insufficiency fractures. Wang et al. [47] used deep
CNNs to detect and classify mandibular fractures on CT scans.
Pranata et al. [48] combined deep learning and Speeded-Up Robust Features (SURF)
for automated calcaneus fracture detection in CT images. Tanzi et al. [49] established a base-
line for designing reliable deep learning approaches for X-ray bone fracture classification.
While there are previously published systemic reviews on the same topic, they differ
from ours in the purpose and questions asked. Particularly, the systematic review and meta-
analysis by Jung et al. [50] focus primarily on deep learning models and their performance
given different type of data types; however, this review expands on these foundations
Diagnostics 2024, 14, 1879 4 of 20

by exploring challenges in the deployment of AI with advanced imaging techniques into

the clinical settings. Additionally, Kuo et al. [51] focused on comparing models and
physicians performances.
The advent of Generative AI and Large Language Models (LLMs) like OpenAI’s
GPT-3 and GPT-4 has expanded AI’s capabilities in medical imaging. These models can
process and interpret extensive medical literature and imaging data, offering comprehen-
sive support for diagnostic and treatment decisions [52,53]. LLMs assist radiologists by
summarizing findings, suggesting differential diagnoses, and predicting complications
based on imaging data and patient history [54,55]. The integration of LLMs with imaging
AI provides a holistic approach to patient care, combining advanced imaging techniques
with the analytical power of language models [56,57].
In summary, the integration of AI into medical imaging, particularly for bone fracture
detection, has shown significant promise across various studies. Continued development of
large datasets, advanced algorithms, and systematic review processes will further enhance
the efficacy and reliability of AI technologies in medical diagnostics.

3. Methodology
This review adheres to the guidelines established by Okoli and Schabram [17], which
encompass the following eight steps:
• Determine the purpose and formulate research questions: The initial step involves
defining the purpose and formulating the research questions for the review. This
step is essential to provide clarity for readers and to enhance the efficiency of the
review process.
• Draft a detailed protocol: The second step is to draft a comprehensive protocol for the
review and ensure all reviewers are trained to follow it. This guarantees that everyone
involved is aligned on the review procedures.
• Conduct a search for relevant articles: In the third step, a thorough search for rel-
evant articles is conducted. Modern literature searches typically utilize electronic
databases such as IEEE Xplore and ProQuest. Reviewers must be adept at using
Boolean operators to perform effective searches in these databases.
• Screen articles for inclusion: The fourth step involves screening the identified articles
for inclusion in the review. Reviewers must decide which articles to include or exclude
and provide justifications for these decisions.
• Assess the quality of articles: In the fifth step, reviewers assess the quality of the
selected articles. It is crucial to include only high-quality articles, as the overall quality
of the review depends significantly on this.
• Extract data from included articles: The sixth step involves extracting relevant data
from the included articles. These data will be used in the subsequent stages of the
review process.
• Data synthesis: The seventh step, known as data synthesis, involves analyzing the
extracted data. This process may include aggregating, organizing, comparing, and
discussing the findings from the articles.
• Write the systematic review: the final step is to write the systematic review, following
standard research writing principles and ensuring the review is detailed enough for
others to reproduce its results.
The following sections provide detailed explanations of how these eight steps were
implemented in this systematic review.

3.1. Research Questions

Based on the purpose of this study, the following research questions were formulated:
• How effective are Artificial Intelligence techniques in detecting bone fractures?
• What benefits do Artificial Intelligence techniques offer for bone fracture detection?
• What applications have been developed using AI for bone fracture detection?
Diagnostics 2024, 14, 1879 5 of 20

• What challenges do these applications face in clinical settings?

3.2. Research Objectives

This study aims to review and summarize the current research on the use of ma-
chine learning in medical imaging, focusing on its effectiveness, challenges, and potential
research areas.

3.3. Search Approach

To collect relevant articles for this systematic review, we searched nine major scientific
databases: Academic Search Complete, Applied Science & Technology Source, Springer
Nature Journals, ScienceDirect, Journals@OVID, Directory of Open Access Journals, Radio-
logical Society of North America, MEDLINE, and JSTOR Journals. These databases were
selected based on recommendations from the Monash University Library website [58] and
our own experience. These databases are known for indexing high-impact, high-quality
articles in healthcare and information technology. The last search was conducted at the end
of September 2023.

3.4. Inclusion Criteria

• The study uses an AI model for bone fracture detection.
• The study evaluates the performance of the AI model used.
• The dataset used to evaluate the model is well described.
• The study is written in English.

3.5. Exclusion Criteria

• The full text of the article is not available online.
• The article is in the form of a poster, tutorial, abstract, or presentation.
• The article is not in English.
• The study does not evaluate the performance of the AI model used.
• The dataset used to evaluate the model is not well described.

3.6. Keywords
This section presents the keywords used in each database to search for relevant articles.
Refer to Table 1 for details.

Table 1. Table contains a list of keywords used to search for articles in each database.

Database Query Strings

(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“X-ray”
Academic Search Complete
AND “Diagnostic Performance”) OR (“CT scans”)
(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“Machine
Applied Science & Technology Source
Learning” OR “Deep Learning” OR “Imaging”)
(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“Machine
Springer Nature Journals
Learning” OR “Deep Learning” OR “Imaging”)
(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“Machine
ScienceDirect
Learning”)
Journals@OVID (“Fracture Diagnosis” AND “Imaging” AND Bone Fracture)
Directory of Open Access Journals (“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”)
(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“Machine
Radiological Society of North America
Learning”)
(“Bone Fracture” OR “Fracture”) AND (“Detection” OR “Diagnosis”) AND (“X-ray”
MEDLINE
AND “Diagnostic Performance”) AND (“CT scans”)
JSTOR Journals (“Fracture Diagnosis” AND “Imaging” AND Bone Fracture)
Diagnostics 2024, 14, 1879 6 of 20

3.7. Data Extraction

The elements extracted from each article include: Title, Author(s), Type (e.g., Journal
Article, Conference, Workshop), Date of Publication, Country of Origin, Study Design,
Sample Size, AI Technique Used, Performance Metrics (e.g., accuracy, sensitivity, speci-
ficity, AUC), Comparison with Other Methods, Validation Methods, Statistical Significance,
Efficiency Improvements, Accuracy Improvements, Cost-Effectiveness, Accessibility, En-
hanced Diagnostic Capabilities, Software Applications, Commercial Products, Integration
with Medical Systems, Use Cases, Technical Challenges, Regulatory and Ethical Issues,
User Acceptance, Integration Challenges, Data Privacy and Security, Cost and Resource
Requirements, Future Research Areas, and Comments on the quality of the work and any
limitations. Detailed information is provided in Table 2.

Table 2. This table contains the data we extracted from each paper selected in this review study. Some
data were not found in every paper.

Data Item Description

Title Title of the paper
Author(s) Author name(s)
Type e.g., conference/workshop/journal
Date Publishing year
Country Country of authors
Study Design Type of study (e.g., retrospective, prospective)
Sample Size Number of subjects or images analyzed
AI Technique Used Specific AI methods (e.g., deep learning, machine learning algorithms)
Performance Metrics Accuracy, sensitivity, specificity, precision, recall, F1 score, ROC-AUC
Comparison with Other Methods Performance compared to traditional methods or other AI techniques
Validation Methods Cross-validation, external validation, or use of independent test sets
Statistical Significance P-values or confidence intervals to determine significance
Efficiency Improvements Speed of detection, time savings for radiologists
Accuracy Improvements Increased diagnostic accuracy, reduction in human error
Cost-Effectiveness Cost savings in the diagnostic process
Accessibility Improved access to diagnostics in remote or underserved areas
Enhanced Diagnostic Capabilities Detection of fractures that are difficult to identify with traditional methods
Software Applications Names and descriptions of software developed
Commercial Products AI-based products available on the market
Integration with Medical Systems How applications integrate with existing medical imaging systems (e.g., PACS, RIS)
Use Cases Examples of applications in clinical practice
Technical Challenges Issues related to algorithm performance, data quality, and computational requirements
Regulatory and Ethical Issues Regulatory hurdles, ethical concerns regarding AI use in healthcare
User Acceptance Acceptance and trust by medical professionals
Integration Challenges Difficulty in integrating AI applications with existing hospital systems
Data Privacy and Security Concerns about patient data privacy and cybersecurity
Cost and Resource Requirements Costs of implementation, need for specialized resources or training
Future Areas Areas of future investigation
Comments Remarks about the quality of the paper
Diagnostics 2024, 14, 1879 7 of 20

Table 3 provides a comprehensive comparison of various studies on AI-based fracture

detection across different skeletal joints. The table summarizes the dataset characteristics,
type of images used, models applied, skeletal joints targeted, study descriptions, perfor-
mance metrics, and key remarks for each study. It includes diverse datasets ranging from
radiographs to CT scans, and employs a variety of AI models such as AlexNet, VGG, Incep-
tion V3, ResNet, and DenseNet-121. These models are applied to different skeletal parts
including the wrist, hand, ankle, humerus, hips, elbow, chest, vertical roots, skull, ankle,
femur, and mandibule. The table highlights the advancements in AI for medical imaging,
demonstrating high performance metrics such as accuracy, sensitivity, specificity, and AUC
across studies. For instance, Olczak et al. [59] showed that deep learning can exceed human
performance in fracture detection with an accuracy of 0.83, while Kim et al. [32] validated
the use of transfer learning from CNNs with an AUC of 95.4. This comparative analysis un-
derscores the potential of AI to enhance fracture detection accuracy and efficiency, serving
as a valuable resource for further research and application in medical diagnostics.

Table 3. Comparison of various studies on AI-based fracture detection across different skeletal joints.
The table summarizes the dataset characteristics, type of images used, models applied, skeletal joints
targeted, study descriptions, performance metrics, and key remarks for each study.

No. Reference Year Dataset Modality Model Parts Description Performance

1 Olczak 2017 The dataset consists of Radiographic AlexNet, Various This research Accuracy = 0.83
et al. [59] 256,000 radiographs of images NIN, Parts demonstrates that
the wrist, hand, and an- VGGs deep learning can
kle. exceed human
performance.
2 Kim 2018 There are 695 wrist radio- Radiographic Inception Wrist The author vali- AUC = 95.4,
et al. [32] graphs with fractures and images V3 dated that using Sensitivity =
694 without fractures. transfer learning 90, Specificity
from CNNs for = 88
fracture detection
on radiographs
can achieve top-
tier performance.
3 Chung 2018 The dataset includes 1891 Radiographic Resnet 152 Humeral The authors in- Accuracy =
et al. [33] plain shoulder AP radio- images troduced a model 96, Sensitiv-
graphs (1376 with prox- for identifying ity = 0.99,
imal humerus fractures and categorizing Specificity =
and 515 normal shoul- fractures from AP 0.97, AUC =
ders) from 1891 patients, shoulder radio- 0.996
comprising 591 men and graphic images.
1300 women.
4 Urakawa 2018 The dataset includes 3346 Radiographic VGG16 Hips This study Accuracy =
et al. [34] hip images, with 1773 images presents a perfor- 95.5, Sensi-
fractured and 1573 non- mance comparison tivity = 93.9,
fractured, collected from between CNNs Specificity =
the Department of Ortho- and orthopedic 97.40, AUC =
pedic Surgery, Tsuruo. surgeons. 0.984
5 Yu 2019 There are 307 patients Radiographic InceptionV3 Hips The proposed Accuracy =
et al. [35] with APFFs and 310 nor- images algorithm excelled 96.9, AUC =
mal patients. in detecting APFF 0.994, Sensi-
but struggled with tivity = 97.1,
precise fracture Specificity =
localization. 96.7
Diagnostics 2024, 14, 1879 8 of 20

Table 3. Cont.

No. Reference Year Dataset Modality Model Parts Description Performance

6 Choi 2019 The dataset comprises Radiographic Resnet 50 Elbow The authors AUC = 0.985,
et al. [36] 1266 pairs of AP and lat- images focused on devel- Sensitiv-
eral elbow radiographs oping a dual-input ity = 93.9,
examined between Jan- CNN-based model Specificity =
uary 2013 and December for the automated 92.2
2017 at a single institu- detection of supra-
tion. condylar fractures.

7 Majkowska 2020 The study utilized two Radiographic Xception Chest The authors created AUC = 0.86,
et al. [37] datasets: DS1 with images a model to identify Sensitiv-
759,611 images from a opacity, pneumoth- ity = 59.9,
multicity hospital net- orax, mass or nod- Specificity =
work, and ChestX-ray14, ule, and fractures. 99.4
a publicly available
dataset with 112,120 im-
ages. Natural language
processing and expert
review labeled 657,954
training images. The
test sets included 1818
images from DS1 and
1962 from ChestX-ray14.

8 Johari 2016 The dataset contains 240 Radiographic Probabilistic Vertical This study sup- Accuracy =
et al. [38] radiographs of teeth: images neural Roots ports the prelim- 96.6, Sensi-
120 with no VRFs and network inary detection tivity = 93.3,
120 with vertical frac- (PNN) of vertical root Specificity =
tures. Each category is CBCT- fractures. 100, CMPIs
split equally between G1/2/3, THRESH-
endodontically treated PA- OLD = 0.79
and untreated teeth. G1/2/3

9 Heimer 2018 The dataset includes 84 Images are Deep Skull The study aims Specificity
et al. [39] skull fracture cases, with extracted neural to classify and = 87.5, Sen-
5 excluded due to se- from Post- networks detect skull frac- sitivity =
vere destruction and 4 re- mortem tures using curved 91.4, CMPIs
moved because of surgi- computed maximum intensity THRESH-
cal material. For each of tomog- projections (CMIP) OLD = 0.75
the 75 included cases, a raphy and deep neural
corresponding case with- (PMCT) networks.
out documented skull
fractures was retrieved.

10 Kitamura 2019 There are 298 normal and Radiographic Seven Ankle The study evalu- Best perfor-
et al. [40] 298 fractured ankle stud- images models: ated the efficiency mance by
ies identified by parsing Incep- of CNNs on small Ensemble_A,
radiology reports. tion V3, datasets. Accuracy =
ResNet, 83, Sensi-
Xception tivity = 80,
Specificity =
81
Diagnostics 2024, 14, 1879 9 of 20

Table 3. Cont.

No. Reference Year Dataset Modality Model Parts Description Performance

11 Gan 2019 The training dataset Radiographic Inception Wrist The authors im- Accuracy =
et al. [41] includes 2040 images images V4 plemented an al- 93, AUC =
(1341 with DRFs and gorithm to de- 0.961, Sensi-
699 without DRFs) and tect distal radius tivity = 90,
a test dataset of 300 fractures. Specificity =
images (150 with DRFs 96
and 150 without DRFs).

12 Lindsey 2018 The dataset comprises Radiographic Unet Wrist This study in- AUC =
et al. [42] 135,845 radiographs images volves using 97.5%, Sen-
of various body parts. deep learning sitivity
The remaining 100,855 to help doctors = 93.9%,
radiographs cover distinguish be- Specificity =
11 body parts. The tween fractured 94.5%
shoulder has the most and normal
radiographs (26,042), wrists.
while the spine has the
least (885).

13 Adams 2019 Various dataset sizes Radiographic AlexNet Femur The author Accuracy
et al. [43] (200, 320, and 640 im- images and aimed to assess AlexNet
ages) are split into train- GoogLeNet the accuracy = 89.4%,
ing (80%) and valida- of DCNN for GoogLeNet
tion (20%), with an addi- femur fracture = 94.4%
tional 160 images used detection.
as the final test set.

14 Rahmaniar 2018 The dataset includes A com- Computerized Calcaneal In this study, the Accuracy =
et al. [44] 815 coronal images, 777 puterized system Frac- author focused 0.86, preci-
transverse images, and tomogra- tures on detecting sion rate =
618 sagittal images. phy (CT) femoral neck 0.86, recall =
images fractures using 0.89
genetic and
deep learning
methods.

15 Tomita 2018 The dataset consists A com- Deep con- vertebra This study aims Accuracy
et al. [45] of 1432 CT scans, puterized volutional at the early de- = 89.2%,
comprising 10,546 two- tomogra- neural tection of osteo- F1 score
dimensional images in phy (CT) network porotic vertebral = 90.8%,
sagittal view, with a test images (CNN) fractures. sensitivity
set of 129 CT scans. = 85.2%,
specificity =
95.8%

16 Muehlematter 2019 Standard CT scans of A com- Machine- vertebra The author AUC = 0.64
et al. [46] 58 patients with insuf- puterized learning aimed to evalu-
ficiency fractures of the tomogra- algorithms ate the perfor-
spine were performed phy (CT) mance of bone
between 2006 and 2013. images texture analysis
with a ma-
chine learning
algorithm.

17 Wang 2022 The dataset includes A com- U-Net and Mandibule The author de- Accuracy =
et al. [47] 222 training images, 56 puterized ResNet veloped a novel 90%, AUC =
validation images, and tomogra- method for clas- 0.956
408 testing images of CT phy (CT) sifying and de-
scans. images tecting mandibu-
lar fractures.
Diagnostics 2024, 14, 1879 10 of 20

Table 3. Cont.

No. Reference Year Dataset Modality Model Parts Description Performance

18 Pranata 2019 Two datasets were used: A com- ResNet Femoral The author aimed Accuracy =
et al. [48] the first contains 255 frac- puterized and VGG Neck at detecting 0.793, Speci-
tured and 732 normal im- tomogra- femoral neck frac- ficity = 0.729,
ages (totaling 987), and phy (CT) tures using genetic Sensitivity =
the second includes 428 images and deep learning 0.829
fractured and 516 normal techniques.
images (totaling 944).

19 Cheng 2019 The dataset comprises Radiographic DenseNet- limb The goal of this accuracy =
et al. [60] 25,505 hip radiographs. images 121 study was to lo- 91%, sen-
calize and classify sitivity =
hip fractures using 98%, false-
deep learning. negative rate
= 2%, AUC =
0.98

3.8. Data Analysis

After extracting data from the papers, the analysis was conducted based on four main
themes derived from the research questions: application, benefits, challenges, and future
areas of application. Each main theme included several sub-themes identified during the
data analysis.

4. Results
This section summarizes the process in four parts: (1) search, (2) exclusion, (3) inclu-
sion, and (4) eligibility, as shown in Figure 1.

Figure 1. Summary of data extraction process.

During the search step, potential articles for the review were identified from various
scientific databases. A total of 33 articles were found from sources such as Academic Search
Complete, Applied Science & Technology Source, Springer Nature Journals, ScienceDirect,
Journals@OVID, Directory of Open Access Journals, Radiological Society of North America,
MEDLINE, and JSTOR Journals. This step utilizes the keywords mentioned in Table 1 to
search for and determine the initial pool of articles considered for the review. The number
of articles found in each database is indicated in Figure 1.
Diagnostics 2024, 14, 1879 11 of 20

The exclusion step involved filtering out articles from the initial pool based on specific
criteria. Fourteen articles were excluded because they were not related to human bone
fractures (e.g., related to animals), unrelated to bone fracture classification and identification,
or were not within the specified date range. This step ensures the review focuses on the
most relevant and timely studies.
In the inclusion step, the remaining articles were further evaluated to determine their
relevance to the review. Nineteen articles met the criteria, and were included for further
analysis in the systematic review. This step forms the foundation of the systematic review,
as these included articles will be thoroughly analyzed.
Finally, in the eligibility step, the full text of these 19 articles was included in the
systematic review after confirming their relevance and quality. This final step ensures that
only the most relevant and high-quality studies are included in the review.

4.1. Publication Year and Geographic Distribution

The selected studies span from 1995 to 2023, with a noticeable increase in publications
in the last decade, reflecting the growing interest and advancements in AI applications for
medical imaging. Most research originates from countries with significant investments
in healthcare and technology, including the United States, China, the United Kingdom,
Japan, and Germany. This distribution highlights a global effort to integrate AI into medical
diagnostics, with notable contributions from both academic institutions and industry
leaders. The rise in AI-related publications aligns with the broader trend of increasing
computational power, improved algorithms, and the availability of large medical imaging
datasets such as MURA [26], ChestX-ray14 [37], and others [27,29].
The distribution of publication years (Table 4) shows a significant increase in research
activity in the last decade, particularly between 2016 and 2020. This trend underscores the
growing interest in applying AI technologies to medical imaging, driven by advances in compu-
tational power and algorithmic improvements. Geographically, the research is predominantly
conducted in countries with substantial investments in healthcare and technology (Table 5),
reflecting their capacity to support cutting-edge research in AI and medical diagnostics.

Table 4. Publication year distribution of selected studies.

Year Range Number of Publications

2016–2017 2
2018–2019 13
2020–2021 1
2022–2023 1

Table 5. Geographic distribution of selected studies.

Country Number of Publications

United States 6
China 4
Japan 3
United Kingdom 2
Iran 2
Switzerland 2
India 1
Australia 1
South Korea 1
Diagnostics 2024, 14, 1879 12 of 20

4.2. Publication Types and Venues

The studies included in this review were published in a variety of high-impact journals
and conferences, indicating the rigorous peer-review processes and the recognition of AI’s
potential in medical imaging. Notable journals include the Journal of Medical Imaging,
Radiology, IEEE Transactions on Medical Imaging, Artificial Intelligence in Medicine, Diagnostics,
and Clinical Radiology. Conferences such as the International Conference on Medical Image
Computing and Computer-Assisted Intervention (MICCAI), IEEE International Symposium on
Biomedical Imaging (ISBI), and SPIE Medical Imaging also feature prominently. These venues
are well-regarded for their focus on cutting-edge research and innovative applications in
the field of medical imaging and AI.
The majority of the selected studies were published in high-impact journals (Table 6),
reflecting the importance and rigor of the research. Conferences also play a critical role, par-
ticularly those focusing on medical imaging and AI. The top journals and conferences listed
in Table 7 underscore the broad interest and recognition of AI’s potential to revolutionize
medical diagnostics.

Table 6. Publication types of selected studies.

Type Number of Publications

Journal 18
Conference 2

Table 7. Top journals and conferences of selected studies.

Journal/Conference Name Number of Publications

Diagnostics 2
Clinical Radiology 2
Acta Orthopaedica 2
Skeletal Radiology 1
Dentomaxillofacial Radiology 1
Computers in Biology and Medicine 1
European Radiology 1
Clinical Oral Investigations 1
Journal of Medical Imaging and Radiation Oncology 1
IEEE Transactions on Medical Imaging 1
Proceedings of the National Academy of Sciences 1
Journal of Medical Imaging 1
Radiology 1
Artificial Intelligence in Medicine 1

5. Discussion
5.1. Effectiveness
The reviewed studies collectively highlight the high effectiveness of AI-based models
in detecting bone fractures. Models like CNN, ResNet, and VGG16 have demonstrated
superior performance metrics, often surpassing human radiologists in terms of accuracy,
sensitivity, and specificity. For instance, the study by Lindsey et al. [42] using a deep neural
network reported a significant improvement in fracture detection rates. Similarly, deep
learning models applied to hip fracture detection achieved impressive results, indicating
the robustness of these technologies in clinical applications [33,34].
Diagnostics 2024, 14, 1879 13 of 20

Additionally, the effectiveness of AI in bone fracture detection is evidenced by its

performance across various anatomical sites. In studies focusing on wrist fractures, AI
models have achieved accuracy levels comparable to expert radiologists, demonstrating
their potential to support clinical decision-making [32]. The application of AI in detecting
rib fractures from chest radiographs also showed high accuracy and reduced time for
diagnosis, further emphasizing the efficiency of these models [37]. Moreover, in a study
by Guo et al. [61], real-time AI-assisted diagnostic systems significantly outperformed
traditional methods in identifying subtle fractures, showcasing the technology’s potential
to enhance diagnostic workflows.
Furthermore, the integration of AI algorithms with advanced imaging techniques such
as 3D CT and MRI has led to even higher diagnostic accuracies. AI’s ability to process and
analyze complex imaging data allows it to identify minute fracture details that might be
overlooked by human eyes. This capability underscores its critical role in modern radiology,
offering a level of precision that enhances diagnostic confidence and accuracy [62].
The enhanced detection capability of AI not only aids in accurate diagnosis but
also significantly contributes to better patient management and treatment outcomes. By
providing detailed and precise imaging analysis, AI assists in formulating more effective
treatment plans, leading to improved patient recovery rates. Moreover, AI-driven imaging
analysis can detect subtle changes over time, which is crucial for monitoring the progression
of conditions and adjusting treatments accordingly.
In particular, studies have demonstrated that AI models are crucial in emergency
settings, where rapid and accurate diagnosis is essential to patient care. The ability to
quickly and accurately diagnose fractures in emergency situations can make a significant
difference in patient outcomes. For instance, in trauma cases where multiple injuries need
to be assessed simultaneously, AI can prioritize and identify critical fractures that require
immediate attention [63–66]. This prioritization is vital for ensuring timely interventions,
which can prevent complications and improve survival rates.
The rise of Large Language Models (LLMs) such as OpenAI’s ChatGPT further en-
hances the integration of AI in medical imaging. These models can process and interpret
vast amounts of medical literature and imaging data, providing comprehensive support for
diagnostic and treatment decisions. LLMs can assist radiologists by summarizing findings,
suggesting differential diagnoses, and even predicting potential complications based on
the imaging data and patient history [67,68]. This integration of LLMs with imaging AI
offers a holistic approach to patient care, combining the strengths of advanced imaging
techniques with the analytical power of language models.
Overall, the integration of AI with advanced imaging technologies represents a sig-
nificant advancement in medical diagnostics. It leverages the strengths of both fields to
provide more accurate, efficient, and comprehensive care, ultimately enhancing patient
outcomes and setting a new standard for radiological practice.

5.2. Benefits
The benefits of using AI techniques in bone fracture detection are manifold:
• Accuracy improvements: AI models provide higher accuracy in detecting fractures
compared to traditional methods. Studies have shown that AI can significantly reduce
the rate of missed fractures, particularly in complex cases where human error is
more likely [37]. This increased accuracy ensures that patients receive timely and
appropriate treatment, reducing the risk of complications.
• Efficiency improvements: Faster analysis and diagnosis, reducing the workload for
radiologists. By automating the initial screening process, AI systems can quickly
identify and flag potential fractures, allowing radiologists to focus on more complex
cases [61,69]. This leads to improved workflow efficiency and shorter patient waiting
times, ultimately enhancing the overall patient experience.
• Enhanced diagnostic capabilities: Improved ability to detect subtle fractures that
might be missed by human eyes. AI algorithms are trained on vast datasets, enabling
Diagnostics 2024, 14, 1879 14 of 20

them to recognize patterns and anomalies that may not be apparent to human ob-
servers [62,70]. This capability is particularly valuable in detecting hairline fractures
and other minor injuries, which can be crucial for early intervention and treatment.
• Accessibility: Enhanced access to diagnostic tools in remote and underserved ar-
eas. AI-powered diagnostic tools can be deployed in regions with limited access to
specialist radiologists, providing high-quality diagnostic support where it is most
needed [71,72]. This democratizes healthcare and ensures that patients in all locations
can benefit from advanced diagnostic technologies.
• Cost-effectiveness: While not always discussed, the potential cost savings from
faster, more accurate diagnoses and reduced need for follow-up imaging and treat-
ments can be significant. AI applications can streamline workflow, reduce overhead
costs, and enhance overall healthcare efficiency [73]. This is particularly important
in resource-limited settings, where cost savings can make advanced diagnostic tech-
niques more accessible.

5.3. Applications
The application of AI in medical imaging has already been deployed in various
settings, and is expected to see even wider adoption in the near future [74]. Several studies
provide concrete examples of AI applications in bone fracture detection, demonstrating the
versatility and effectiveness of these technologies:
• Proximal humerus fracture detection: Studies using a deep learning model known as
ResNet 152 have achieved an impressive accuracy of 96%, highlighting the model’s
effectiveness in identifying shoulder fractures [33]. This application is particularly
beneficial in emergency settings where quick and accurate diagnosis is critical. The
ability to promptly identify proximal humerus fractures can significantly reduce
the time to treatment, thereby improving patient outcomes and reducing the risk
of complications.
• Intertrochanteric hip fracture detection: Another deep learning model, often referred
to as VGG16, has demonstrated high sensitivity and specificity, with reported sensitiv-
ity of 93.9% and specificity of 97.4% [34]. The high performance of these models makes
them suitable for integration into clinical workflows, facilitating early and accurate
detection of hip fractures. This early detection is crucial for timely intervention, which
can enhance recovery rates and reduce the burden on healthcare systems.
• Chest radiograph interpretation: Deep learning models have shown exceptional
performance in detecting various thoracic diseases, including rib fractures [37]. The
ability to accurately interpret chest radiographs is invaluable in the diagnosis of
trauma patients. In emergency and trauma care settings, where timely and accurate
diagnosis can be life-saving, AI-enhanced chest radiograph interpretation ensures that
all injuries are promptly identified and treated.
• Wrist fracture detection: The Inception V3 model, another sophisticated AI tool, has
been effectively utilized to detect wrist fractures, achieving high accuracy metrics [32].
This application is particularly useful in sports medicine and orthopedics, where wrist
injuries are common. By providing quick and accurate diagnoses, AI tools can help
manage and treat sports injuries more effectively, potentially reducing downtime for
athletes and expediting their return to activity.
• Automated reporting systems: AI models integrated into clinical workflows have
significantly improved diagnostic efficiency, as evidenced by various studies [26,43].
These systems can automatically generate detailed reports, reducing the adminis-
trative burden on radiologists and allowing them to focus more on patient care.
Automated reporting also ensures consistency and reduces the risk of human error in
documentation, leading to more reliable and standardized diagnostic outputs.
The rise of Generative AI is expected to accelerate the integration of AI technologies
in clinical settings [53,75–77]. Generative AI can enhance diagnostic processes by creating
realistic synthetic data for training models, generating detailed and tailored diagnostic
Diagnostics 2024, 14, 1879 15 of 20

reports, and even simulating various clinical scenarios to improve decision-making. These
advancements can further streamline clinical workflows and enhance the overall quality of
patient care.
These examples illustrate the broad range of applications for AI in bone fracture
detection and the significant benefits they bring to clinical practice. From improving
diagnostic accuracy and efficiency to enhancing patient outcomes, AI technologies are
poised to revolutionize the field of medical imaging.

5.4. Challenges
Despite the promising results, several challenges remain:
• Data quality and quantity: Effective AI models require high-quality, annotated
datasets, which are often scarce and expensive to produce [37]. Additionally, the
diversity of training data are crucial to ensure AI models perform well across differ-
ent populations and clinical scenarios. Efforts to create and share extensive, diverse
datasets could help address this issue.
• Generalizability: AI models trained on specific datasets might not perform well
with different populations or imaging methods. Ensuring these models are robust
across various clinical settings remains a significant challenge [31,43]. Validating AI
models in diverse clinical environments is necessary to ensure their reliability and
effectiveness. This includes multicenter studies and data from various geographic and
demographic backgrounds.
• Integration into clinical workflows: Integrating AI models into current clinical work-
flows involves overcoming technical, regulatory, and acceptance barriers. It is cru-
cial to ensure smooth integration without disrupting clinical practices [45,46]. This
includes technical integration and training healthcare providers to use AI tools effec-
tively and trust their outputs. Successful integration requires collaboration between
AI developers, healthcare providers, and regulatory authorities.
• Ethical and legal considerations: The use of AI in healthcare raises ethical and legal
issues, including patient privacy, data security, and potential biases in AI algorithms.
Addressing these concerns is essential for the broad adoption of AI technologies in
medical diagnostics [28]. Regulatory frameworks must be established to ensure AI
applications meet legal standards and ethical guidelines. Transparency and account-
ability in AI decision-making processes are also crucial for maintaining trust.
• User acceptance: For AI technologies to be successfully implemented, healthcare pro-
fessionals must accept them. Training and education are necessary to build trust and
confidence in AI-assisted diagnostic tools [78]. Resistance to change and unfamiliarity
with AI technologies can impede adoption. Strategies to increase user acceptance
include demonstrating the reliability and effectiveness of AI tools through clinical
trials and providing ongoing education and support to healthcare providers.
• Technical challenges: Variations in imaging protocols and quality can impact the
performance of AI models. Ensuring consistency in image acquisition and addressing
technical variations are essential for reliable AI performance [33]. Developing robust
AI models capable of handling a wide range of imaging conditions and integrating
quality control measures into AI workflows can help mitigate these challenges.

5.5. Research Contributions

This review makes several significant research contributions:
• Comprehensive synthesis: this review provides a detailed synthesis of data from
various studies, offering a thorough overview of the current state of AI in bone fracture
detection, focusing on its effectiveness, benefits, applications, and challenges.
• Identification of key trends: it highlights key trends in AI research, such as the
growing use of deep learning models like CNN, ResNet, and VGG16, which show
superior performance in various fracture detection tasks.
Diagnostics 2024, 14, 1879 16 of 20

• Emphasis on data quality: the review underscores the essential role of high-quality,
annotated datasets in training effective AI models and the need for initiatives to create
and share extensive, diverse datasets.
• Addressing generalizability issues: by discussing the challenges related to the gen-
eralizability of AI models, the review emphasizes the importance of validating AI
models in diverse clinical settings to ensure their reliability and effectiveness.
• Focus on integration and acceptance: the review highlights the importance of seamlessly
integrating AI into clinical workflows and the need for training and education to build
trust and confidence in AI-assisted diagnostic tools among healthcare professionals.
• Ethical and legal considerations: by addressing ethical and legal considerations, the
review calls for the establishment of regulatory frameworks to ensure that AI applica-
tions comply with legal standards and ethical guidelines, ensuring transparency and
accountability in AI decision-making processes.
• Future research directions: the review suggests future research directions, includ-
ing further validation studies, improving data quality, enhancing model robustness,
facilitating seamless integration into clinical workflows, and addressing ethical and
legal concerns.

5.6. Limitations
While this systematic review provides valuable insights into the applications of AI in
bone fracture detection, several limitations should be acknowledged.
• Database selection: The review was conducted using specific databases recom-
mended for their strong indexing of high-impact, high-quality articles in health-
care and information technology. However, the exclusion of broad multidisciplinary
databases like WoS and Scopus may have led to the omission of some relevant studies.
Future research could benefit from a more comprehensive search strategy that includes
these databases to ensure broader coverage of the literature.
• Search strategy and keywords: The search strategy primarily focused on general AI
and machine learning terms, such as “Artificial Intelligence” and “Machine Learn-
ing,” without including specific algorithms like Neural Networks, Decision Trees,
Random Forests, KNN, and Bayes. This was an intentional decision to capture a broad
spectrum of AI applications rather than narrow the focus to incremental performance
improvements of specific models. However, this approach may have led to the exclu-
sion of studies that focus on the detailed performance of particular machine learning
techniques. Future reviews could consider incorporating these specific terms to ensure
a more exhaustive collection of relevant studies.
• Study heterogeneity: The included studies vary significantly in terms of sample size,
imaging modalities, and AI models used. This heterogeneity makes it challenging
to directly compare results across studies. The lack of standardized reporting on AI
model performance and the diverse clinical settings may also affect the generalizability
of the findings.
• Lack of meta-analysis: Unlike some other systematic reviews, this study did not
perform a meta-analysis due to the variability in study designs and outcome measures.
This limits the ability to quantitatively synthesize the results and draw more definitive
conclusions about the overall effectiveness of AI in bone fracture detection.
• Rapidly evolving field: The field of AI in medical imaging is rapidly evolving, with
new techniques and models being developed continuously. As a result, some of
the findings presented in this review may become outdated as newer studies are
published. Continuous updates to the review are necessary to keep pace with the
latest advancements.

6. Conclusions
This systematic review synthesized the current research on AI-based applications for
bone fracture detection using medical images. The findings highlight the high effectiveness
Diagnostics 2024, 14, 1879 17 of 20

and potential of AI technologies in improving diagnostic accuracy and efficiency across

various fracture types and imaging modalities. Studies have shown that advanced AI
models, such as convolutional neural networks (CNNs) like InceptionNet, VGG16, and
ResNet, often surpass human radiologists in accuracy, sensitivity, and specificity [33,34,42].
These models have been successfully applied to detect and classify proximal humerus
fractures, hip fractures, chest radiographs, and dental fractures [33].
The integration of AI into clinical workflows shows promise in reducing the workload
for radiologists and increasing diagnostic throughput, especially through automated reporting
systems. Additionally, AI applications have effectively localized fracture sites and segmented
bones in 3D imaging modalities, aiding clinicians in diagnosis and treatment planning.
However, several challenges must be addressed to fully realize the benefits of AI in
healthcare. High-quality, annotated datasets are crucial for training effective AI models,
but their availability is often limited, and obtaining such datasets can be costly and time-
consuming [37]. Ensuring the generalizability of AI models across diverse populations and
imaging modalities remains a continuous challenge, as models trained on specific datasets
may not perform well in different clinical settings [31,43].
Integrating AI models into existing clinical workflows requires overcoming technical,
regulatory, and acceptance barriers. Ensuring seamless integration without disrupting
clinical practices is vital for the widespread adoption of AI technologies. Additionally,
ethical and legal considerations, including patient privacy, data security, and potential
biases in AI algorithms, must be addressed to ensure the responsible deployment of AI in
healthcare [79,80].
In conclusion, while AI technologies hold significant promise for advancing medical
diagnostics, particularly in bone fracture detection, concerted efforts are needed to address
existing challenges. Future research should focus on improving data quality, enhancing
model robustness and generalizability, facilitating seamless integration into clinical work-
flows, and addressing ethical and legal concerns. By tackling these issues, we can fully
leverage the potential of AI to transform healthcare and improve patient outcomes.

Funding: This research was funded by the Deputyship for Research and Innovation, Ministry of
Education in Saudi Arabia grant number 7871.
Data Availability Statement: No new data were created or analyzed in this study.
Acknowledgments: The author extends his appreciation to the Deputyship for Research and Innovation,
Ministry of Education in Saudi Arabia for funding this research work through Project Number 7871.
Conflicts of Interest: The author declares no conflicts of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or
in the decision to publish the results.

References
1. Choplin, R.H.; Boehme, J., 2nd; Maynard, C.D. Picture archiving and communication systems: An overview. Radiographics 1992,
12, 127–129. [CrossRef]
2. Mansoori, B.; Erhard, K.K.; Sunshine, J.L. Picture archiving and communication system (PACS) implementation, integration &
benefits in an integrated health system. Acad. Radiol. 2012, 19, 229–235.
3. Alhajeri, M.; Shah, S.G.S. Limitations in and solutions for improving the functionality of picture archiving and communication
system: An exploratory study of PACS professionals’ perspectives. J. Digit. Imaging 2019, 32, 54–67. [CrossRef] [PubMed]
4. Sadik, M. Computer-Assisted Diagnosis for the Interpretation of Bone Scintigraphy. A New Approach to Improve Diagnostic
Accuracy. Ph.D. Thesis, Institute of Medicine, Department of Molecular and Clinical Medicine, Unveristy of Gothenburg,
Gothenburg, Sweden, 2009.
5. Hussain, S.; Mubeen, I.; Ullah, N.; Shah, S.S.U.D.; Khan, B.A.; Zahoor, M.; Ullah, R.; Khan, F.A.; Sultan, M.A. Modern diagnostic
imaging technique applications and risk factors in the medical field: A review. BioMed Res. Int. 2022, 2022, 5164970. [CrossRef]
6. Litjens, G.; Kooi, T.; Bejnordi, B.E.; Setio, A.A.A.; Ciompi, F.; Ghafoorian, M.; Van Der Laak, J.A.; Van Ginneken, B.; Sánchez, C.I.
A survey on deep learning in medical image analysis. Med. Image Anal. 2017, 42, 60–88. [CrossRef]
7. Guermazi, A.; Tannoury, C.; Kompel, A.J.; Murakami, A.M.; Ducarouge, A.; Gillibert, A.; Li, X.; Tournier, A.; Lahoud, Y.; Jarraya,
M.; et al. Improving radiographic fracture recognition performance and efficiency using artificial intelligence. Radiology 2022,
302, 627–636. [CrossRef] [PubMed]
Diagnostics 2024, 14, 1879 18 of 20

8. Boginskis, V.; Zadoroznijs, S.; Cernavska, I.; Beikmane, D.; Sauka, J. Artificial intelligence effectivity in fracture detection. Med.
Perspect. 2023, 3, 68–78. [CrossRef]
9. Tieu, A.; Kroen, E.; Kadish, Y.; Liu, Z.; Patel, N.; Zhou, A.; Yilmaz, A.; Lee, S.; Deyer, T. The Role of Artificial Intelligence in the
Identification and Evaluation of Bone Fractures. Bioengineering 2024, 11, 338. [CrossRef]
10. Bhatnagar, A.; Kekatpure, A.L.; Velagala, V.R.; Kekatpure, A. A Review on the Use of Artificial Intelligence in Fracture Detection.
Cureus 2024, 16, e58364. [CrossRef]
11. Yamashita, R.; Nishio, M.; Do, R.K.G.; Togashi, K. Convolutional neural networks: An overview and application in radiology.
Insights Imaging 2018, 9, 611–629. [CrossRef]
12. Vaz, J.M.; Balaji, S. Convolutional neural networks (CNNs): Concepts and applications in pharmacogenomics. Mol. Divers. 2021,
25, 1569–1584. [CrossRef] [PubMed]
13. Yasaka, K.; Akai, H.; Kunimatsu, A.; Kiryu, S.; Abe, O. Deep learning with convolutional neural network in radiology. Jpn. J.
Radiol. 2018, 36, 257–272. [CrossRef]
14. Oakden-Rayner, L. Exploring large-scale public medical image datasets. Acad. Radiol. 2020, 27, 106–112. [CrossRef]
15. Klein, A.; Warszawski, J.; Hillengaß, J.; Maier-Hein, K.H. Automatic bone segmentation in whole-body CT images. Int. J. Comput.
Assist. Radiol. Surg. 2019, 14, 21–29. [CrossRef]
16. Löffler, M.T.; Sekuboyina, A.; Jacob, A.; Grau, A.L.; Scharr, A.; El Husseini, M.; Kallweit, M.; Zimmer, C.; Baum, T.; Kirschke, J.S.
A vertebral segmentation dataset with fracture grading. Radiol. Artif. Intell. 2020, 2, e190138. [CrossRef]
17. Okoli, C.; Schabram, K. A Guide to Conducting a Systematic Literature Review of Information Systems Research. Sprouts Work.
Pap. Inf. Syst. 2010, 10, 1–49. [CrossRef]
18. Dickersin, K.; Scherer, R.; Lefebvre, C. Systematic reviews: Identifying relevant studies for systematic reviews. BMJ 1994,
309, 1286–1291. [CrossRef] [PubMed]
19. Rajpurkar, P.; Irvin, J.; Zhu, K.; Yang, B.; Mehta, H.; Duan, T.; Ding, D.; Bagul, A.; Langlotz, C.; Shpanskaya, K.; et al. Chexnet:
Radiologist-level pneumonia detection on chest x-rays with deep learning. arXiv 2017, arXiv:1711.05225.
20. Lin, R.H. An intelligent model for liver disease diagnosis. Artif. Intell. Med. 2009, 47, 53–62. [CrossRef]
21. Dombi, G.W.; Nandi, P.; Saxe, J.M.; Ledgerwood, A.M.; Lucas, C.E. Prediction of rib fracture injury outcome by an artificial neural
network. J. Trauma Acute Care Surg. 1995, 39, 915–921. [CrossRef]
22. Zhang, Z.; Chen, P.; Sapkota, M.; Yang, L. Tandemnet: Distilling knowledge from medical images using diagnostic reports as
optional semantic references. In Proceedings of the Medical Image Computing and Computer Assisted Intervention—MICCAI 2017: 20th
International Conference, Quebec City, QC, Canada, 11–13 September 2017; Proceedings, Part III 20; Springer: Cham, Switzerland,
2017; pp. 320–328.
23. Ypsilantis, P.P.; Montana, G. Learning what to look in chest X-rays with a recurrent visual attention model. arXiv 2017,
arXiv:1701.06452.
24. Fu, Y.; Liu, R.; Liu, Y.; Lu, J. Intertrochanteric fracture visualization and analysis using a map projection technique. Med. Biol. Eng.
Comput. 2019, 57, 633–642. [CrossRef] [PubMed]
25. Yaqub, M.; Kelly, B.; Papageorghiou, A.T.; Noble, J.A. Guided random forests for identification of key fetal anatomy and image
categorization in ultrasound scans. In Proceedings of the Medical Image Computing and Computer-Assisted Intervention—MICCAI
2015: 18th International Conference, Munich, Germany, 5–9 October 2015; Proceedings, Part III 18; Springer: Cham, Switzerland, 2015;
pp. 687–694.
26. Rajpurkar, P.; Irvin, J.; Bagul, A.; Ding, D.; Duan, T.; Mehta, H.; Yang, B.; Zhu, K.; Laird, D.; Ball, R.L.; et al. Mura: Large dataset
for abnormality detection in musculoskeletal radiographs. arXiv 2017, arXiv:1712.06957.
27. Alammary, A. Blended learning models for introductory programming courses: A systematic review. PLoS ONE 2019, 14, e0221765.
[CrossRef]
28. Liang, W.; Ji, N. Privacy challenges of IoT-based blockchain: A systematic review. Clust. Comput. 2022, 25, 2203–2221. [CrossRef]
29. Konttila, J.; Siira, H.; Kyngäs, H.; Lahtinen, M.; Elo, S.; Kääriäinen, M.; Kaakinen, P.; Oikarinen, A.; Yamakawa, M.; Fukui, S.; et al.
Healthcare professionals’ competence in digitalisation: A systematic review. J. Clin. Nurs. 2019, 28, 745–761. [CrossRef]
30. AlShamsi, M.; Al-Emran, M.; Shaalan, K. A systematic review on blockchain adoption. Appl. Sci. 2022, 12, 4245. [CrossRef]
31. Meena, T.; Roy, S. Bone fracture detection using deep supervised learning from radiological images: A paradigm shift. Diagnostics
2022, 12, 2420. [CrossRef]
32. Kim, D.; MacKinnon, T. Artificial intelligence in fracture detection: Transfer learning from deep convolutional neural networks.
Clin. Radiol. 2018, 73, 439–445. [CrossRef]
33. Chung, S.W.; Han, S.S.; Lee, J.W.; Oh, K.S.; Kim, N.R.; Yoon, J.P.; Kim, J.Y.; Moon, S.H.; Kwon, J.; Lee, H.J.; et al. Automated
detection and classification of the proximal humerus fracture by using deep learning algorithm. Acta Orthop. 2018, 89, 468–473.
[CrossRef]
34. Urakawa, T.; Tanaka, Y.; Goto, S.; Matsuzawa, H.; Watanabe, K.; Endo, N. Detecting intertrochanteric hip fractures with
orthopedist-level accuracy using a deep convolutional neural network. Skelet. Radiol. 2019, 48, 239–244. [CrossRef]
35. Yu, J.; Yu, S.; Erdal, B.; Demirer, M.; Gupta, V.; Bigelow, M.; Salvador, A.; Rink, T.; Lenobel, S.; Prevedello, L.; et al. Detection
and localisation of hip fractures on anteroposterior radiographs with artificial intelligence: Proof of concept. Clin. Radiol. 2020,
75, 237.e1–237.e9. [CrossRef] [PubMed]
Diagnostics 2024, 14, 1879 19 of 20

36. Choi, J.W.; Cho, Y.J.; Lee, S.; Lee, J.; Lee, S.; Choi, Y.H.; Cheon, J.E.; Ha, J.Y. Using a dual-input convolutional neural network
for automated detection of pediatric supracondylar fracture on conventional radiography. Investig. Radiol. 2020, 55, 101–110.
[CrossRef]
37. Majkowska, A.; Mittal, S.; Steiner, D.F.; Reicher, J.J.; McKinney, S.M.; Duggan, G.E.; Eswaran, K.; Cameron Chen, P.H.; Liu, Y.;
Kalidindi, S.R.; et al. Chest radiograph interpretation with deep learning models: Assessment with radiologist-adjudicated
reference standards and population-adjusted evaluation. Radiology 2020, 294, 421–431. [CrossRef]
38. Johari, M.; Esmaeili, F.; Andalib, A.; Garjani, S.; Saberkari, H. Detection of vertical root fractures in intact and endodontically
treated premolar teeth by designing a probabilistic neural network: An ex vivo study. Dentomaxillofacial Radiol. 2017, 46, 20160107.
[CrossRef]
39. Heimer, J.; Thali, M.J.; Ebert, L. Classification based on the presence of skull fractures on curved maximum intensity skull
projections by means of deep learning. J. Forensic Radiol. Imaging 2018, 14, 16–20. [CrossRef]
40. Kitamura, G.; Chung, C.Y.; Moore, B.E. Ankle fracture detection utilizing a convolutional neural network ensemble implemented
with a small sample, de novo training, and multiview incorporation. J. Digit. Imaging 2019, 32, 672–677. [CrossRef]
41. Gan, K.; Xu, D.; Lin, Y.; Shen, Y.; Zhang, T.; Hu, K.; Zhou, K.; Bi, M.; Pan, L.; Wu, W.; et al. Artificial intelligence detection of distal
radius fractures: A comparison between the convolutional neural network and professional assessments. Acta Orthop. 2019,
90, 394–400. [CrossRef]
42. Lindsey, R.; Daluiski, A.; Chopra, S.; Lachapelle, A.; Mozer, M.; Sicular, S.; Hanel, D.; Gardner, M.; Gupta, A.; Hotchkiss, R.; et al.
Deep neural network improves fracture detection by clinicians. Proc. Natl. Acad. Sci. USA 2018, 115, 11591–11596. [CrossRef]
[PubMed]
43. Adams, M.; Chen, W.; Holcdorf, D.; McCusker, M.W.; Howe, P.D.; Gaillard, F. Computer vs human: Deep learning versus
perceptual training for the detection of neck of femur fractures. J. Med. Imaging Radiat. Oncol. 2019, 63, 27–32. [CrossRef]
[PubMed]
44. Rahmaniar, W.; Wang, W.J. Real-time automated segmentation and classification of calcaneal fractures in CT images. Appl. Sci.
2019, 9, 3011. [CrossRef]
45. Tomita, N.; Cheung, Y.Y.; Hassanpour, S. Deep neural networks for automatic detection of osteoporotic vertebral fractures on CT
scans. Comput. Biol. Med. 2018, 98, 8–15. [CrossRef]
46. Muehlematter, U.J.; Mannil, M.; Becker, A.S.; Vokinger, K.N.; Finkenstaedt, T.; Osterhoff, G.; Fischer, M.A.; Guggenberger, R.
Vertebral body insufficiency fractures: Detection of vertebrae at risk on standard CT images using texture analysis and machine
learning. Eur. Radiol. 2019, 29, 2207–2217. [CrossRef]
47. Wang, X.; Xu, Z.; Tong, Y.; Xia, L.; Jie, B.; Ding, P.; Bai, H.; Zhang, Y.; He, Y. Detection and classification of mandibular fracture on
CT scan using deep convolutional neural network. Clin. Oral Investig. 2022, 26, 4593–4601. [CrossRef]
48. Pranata, Y.D.; Wang, K.C.; Wang, J.C.; Idram, I.; Lai, J.Y.; Liu, J.W.; Hsieh, I.H. Deep learning and SURF for automated classification
and detection of calcaneus fractures in CT images. Comput. Methods Programs Biomed. 2019, 171, 27–37. [CrossRef] [PubMed]
49. Tanzi, L.; Vezzetti, E.; Moreno, R.; Moos, S. X-ray bone fracture classification using deep learning: A baseline for designing a
reliable approach. Appl. Sci. 2020, 10, 1507. [CrossRef]
50. Jung, J.; Dai, J.; Liu, B.; Wu, Q. Artificial intelligence in fracture detection with different image modalities and data types: A
systematic review and meta-analysis. PLoS Digit. Health 2024, 3, e0000438. [CrossRef] [PubMed]
51. Kuo, R.Y.; Harrison, C.; Curran, T.A.; Jones, B.; Freethy, A.; Cussons, D.; Stewart, M.; Collins, G.S.; Furniss, D. Artificial intelligence
in fracture detection: A systematic review and meta-analysis. Radiology 2022, 304, 50–62. [CrossRef] [PubMed]
52. Chiesa-Estomba, C.M.; Lechien, J.R.; Vaira, L.A.; Brunet, A.; Cammaroto, G.; Mayo-Yanez, M.; Sanchez-Barrueco, A.; Saga-
Gutierrez, C. Exploring the potential of Chat-GPT as a supportive tool for sialendoscopy clinical decision making and patient
information support. Eur. Arch. Oto-Rhino-Laryngol. 2024, 281, 2081–2086. [CrossRef]
53. Haupt, C.E.; Marks, M. AI-generated medical advice—GPT and beyond. JAMA 2023, 329, 1349–1350. [CrossRef]
54. Ueda, D.; Mitsuyama, Y.; Takita, H.; Horiuchi, D.; Walston, S.L.; Tatekawa, H.; Miki, Y. Diagnostic Performance of ChatGPT from
Patient History and Imaging Findings on the Diagnosis Please Quizzes. Radiology 2023, 308, e231040. [CrossRef] [PubMed]
55. Currie, G.; Robbie, S.; Tually, P. ChatGPT and patient information in nuclear medicine: GPT-3.5 versus GPT-4. J. Nucl. Med.
Technol. 2023, 51, 307–313. [CrossRef]
56. Sievert, M.; Aubreville, M.; Mueller, S.K.; Eckstein, M.; Breininger, K.; Iro, H.; Goncalves, M. Diagnosis of malignancy in
oropharyngeal confocal laser endomicroscopy using GPT 4.0 with vision. Eur. Arch. Oto-Rhino-Laryngol. 2024, 281, 2115–2122.
[CrossRef]
57. Yang, Z.; Yao, Z.; Tasmin, M.; Vashisht, P.; Jang, W.S.; Ouyang, F.; Wang, B.; Berlowitz, D.; Yu, H. Performance of multimodal
gpt-4v on usmle with image: Potential for imaging diagnostic support with explanations. medRxiv 2023, 2023-10.
58. Subject Guides: Databases by Subject: HOME—guides.lib.monash.edu. Available online: https://guides.lib.monash.edu/
subject-databases (accessed on 27 September 2023).
59. Olczak, J.; Fahlberg, N.; Maki, A.; Razavian, A.S.; Jilert, A.; Stark, A.; Sköldenberg, O.; Gordon, M. Artificial intelligence for
analyzing orthopedic trauma radiographs: Deep learning algorithms—Are they on par with humans for diagnosing fractures?
Acta Orthop. 2017, 88, 581–586. [CrossRef]
60. Cheng, C.T.; Ho, T.Y.; Lee, T.Y.; Chang, C.C.; Chou, C.C.; Chen, C.C.; Chung, I.; Liao, C.H. Application of a deep learning
algorithm for detection and visualization of hip fractures on plain pelvic radiographs. Eur. Radiol. 2019, 29, 5469–5477. [CrossRef]
Diagnostics 2024, 14, 1879 20 of 20

61. Guo, L.; Xiao, X.; Wu, C.; Zeng, X.; Zhang, Y.; Du, J.; Bai, S.; Xie, J.; Zhang, Z.; Li, Y.; et al. Real-time automated diagnosis of
precancerous lesions and early esophageal squamous cell carcinoma using a deep learning model (with videos). Gastrointest.
Endosc. 2020, 91, 41–51. [CrossRef] [PubMed]
62. Venigandla, K.; Tatikonda, V.M. Improving Diagnostic Imaging Analysis with RPA and Deep Learning Technologies. Power Syst.
Technol. 2021, 45, 53–60.
63. Boonstra, A.; Laven, M. Influence of artificial intelligence on the work design of emergency department clinicians a systematic
literature review. BMC Health Serv. Res. 2022, 22, 669. [CrossRef]
64. Liu, N.; Zhang, Z.; Ho, A.F.W.; Ong, M.E.H. Artificial intelligence in emergency medicine. J. Emerg. Crit. Care Med. 2018, 2, 82.
[CrossRef]
65. Al-Dasuqi, K.; Johnson, M.H.; Cavallo, J.J. Use of artificial intelligence in emergency radiology: An overview of current
applications, challenges, and opportunities. Clin. Imaging 2022, 89, 61–67. [CrossRef]
66. Hosseini, M.M.; Hosseini, S.T.M.; Qayumi, K.; Ahmady, S.; Koohestani, H.R. The Aspects of Running Artificial Intelligence in
Emergency Care; a Scoping Review. Arch. Acad. Emerg. Med. 2023, 11, e38.
67. Williams, C.Y.; Zack, T.; Miao, B.Y.; Sushil, M.; Wang, M.; Kornblith, A.E.; Butte, A.J. Use of a large language model to assess
clinical acuity of adults in the emergency department. JAMA Netw. Open 2024, 7, e248895. [CrossRef] [PubMed]
68. Petrella, R.J. The AI future of emergency medicine. Ann. Emerg. Med. 2024, 84, 139–153. [CrossRef] [PubMed]
69. Srivastav, S.; Chandrakar, R.; Gupta, S.; Babhulkar, V.; Agrawal, S.; Jaiswal, A.; Prasad, R.; Wanjari, M.B.; Agarwal, S.; Wanjari,
M. ChatGPT in radiology: The advantages and limitations of artificial intelligence for medical imaging diagnosis. Cureus 2023,
15, e41435. [CrossRef] [PubMed]
70. Mijwil, M.M.; Al-Mistarehi, A.H.; Abotaleb, M.; El-kenawy, E.S.M.; Ibrahim, A.; Abdelhamid, A.A.; Eid, M.M. From Pixels to
Diagnoses: Deep Learning’s Impact on Medical Image Processing—A Survey. Wasit J. Comput. Math. Sci. 2023, 2, 9–15. [CrossRef]
71. Yu, K.; Tan, L.; Lin, L.; Cheng, X.; Yi, Z.; Sato, T. Deep-learning-empowered breast cancer auxiliary diagnosis for 5GB remote
E-health. IEEE Wirel. Commun. 2021, 28, 54–61. [CrossRef]
72. Behar, J.A.; Liu, C.; Kotzen, K.; Tsutsui, K.; Corino, V.D.; Singh, J.; Pimentel, M.A.; Warrick, P.; Zaunseder, S.; Andreotti, F.; et al.
Remote health diagnosis and monitoring in the time of COVID-19. Physiol. Meas. 2020, 41, 10TR01. [CrossRef]
73. Schwendicke, F.; Rossi, J.G.; Gostemeyer, G.; Elhennawy, K.; Cantu, A.G.; Gaudin, R.; Chaurasia, A.; Gehrung, S.; Krois, J.
Cost-effectiveness of Artificial Intelligence for Proximal Caries Detection. J. Dent. Res. 2021, 100, 369–376. [CrossRef]
74. Mello-Thoms, C.; Mello, C.A. Clinical applications of artificial intelligence in radiology. Br. J. Radiol. 2023, 96, 20221031. [CrossRef]
75. Waisberg, E.; Ong, J.; Masalkhi, M.; Kamran, S.A.; Zaman, N.; Sarker, P.; Lee, A.G.; Tavakkoli, A. GPT-4: A new era of artificial
intelligence in medicine. Ir. J. Med. Sci. 2023, 192, 3197–3200. [CrossRef]
76. Potočnik, J.; Foley, S.; Thomas, E. Current and potential applications of artificial intelligence in medical imaging practice: A
narrative review. J. Med. Imaging Radiat. Sci. 2023, 54, 76–385. [CrossRef] [PubMed]
77. Lecler, A.; Duron, L.; Soyer, P. Revolutionizing radiology with GPT-based models: Current applications, future possibilities and
limitations of ChatGPT. Diagn. Interv. Imaging 2023, 104, 269–274. [CrossRef]
78. Lambert, S.I.; Madi, M.; Sopka, S.; Lenes, A.; Stange, H.; Buszello, C.P.; Stephan, A. An integrative review on the acceptance of
artificial intelligence among healthcare professionals in hospitals. NPJ Digit. Med. 2023, 6, 111. [CrossRef]
79. Chan, H.P.; Hadjiiski, L.M.; Samala, R.K. Computer-aided diagnosis in the era of deep learning. Med. Phys. 2020, 47, e218–e227.
[CrossRef]
80. Shalaby, A.; Soliman, A.; Elaskary, S.; Refaey, A.; Abdelazim, M.; Khalifa, F. Artificial intelligence based computer-aided diagnosis
applications for brain disorders from medical imaging data. Front. Neurosci. 2023, 17, 998818. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.