KEMBAR78
CETA Conditions Workflow | PDF | Theoretical Computer Science | Data Mining
Scribd
Upload
19 views8 pages

CETA Conditions Workflow

The document describes the workflow of RE:IN Cloud for feature extraction from images and tabular/time series data. It involves uploading data, running it through feature extraction APIs, and downloading the results. It also discusses current and future implementations for extracting cause features from medical images, including texture and intensity features. Structural similarity is measured using t-SNE to reduce dimensions followed by k-means clustering, but having only two clusters limits identifying highly similar images.

Uploaded by

Budi Kurniawan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
19 views8 pages

CETA Conditions Workflow

The document describes the workflow of RE:IN Cloud for feature extraction from images and tabular/time series data. It involves uploading data, running it through feature extraction APIs, and downloading the results. It also discusses current and future implementations for extracting cause features from medical images, including texture and intensity features. Structural similarity is measured using t-SNE to reduce dimensions followed by k-means clustering, but having only two clusters limits identifying highly similar images.

Uploaded by

Budi Kurniawan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 8

Workflow of RE:IN Cloud

Feature
Extraction
(Pyfeats) Feature Values
(for Edge)

Upload Ydata Feature Values


Image Fastdup
the file Profilling (for Re:In)

Tabular
Data/
Timeseries
POST Fastdup & Ydata-Profiling API
POST Pyfeats API
GET Pyfeats API (Download Result File)
Cause Features
Cause Cause_Code V1 V2

Current Implementation: FOS_Mean C1 <=125 >125


FOS_Skewness C2 <=0 >0
• Included 3 effects (Data value features) such as Intensity, FOS_Kurtosis C3 <=0 >0
Complexity and Uniformity (Homogeneity). FOS_Entropy C4 <=0.5 >0.5
• The threshold (limits) are derived under the following condition FOS_CoefficientOfVariat
 Gray scale image with pixel range (0 to 255) ion C5 <1 >1
 Calculation based on 5x5 pixels.
GLCM_Entropy C6 <=0.5 >0.5
GLCM_ASM C7 <=0.5 >0.5
Drawbacks: GLCM_Contrast C8 <=128 >128
• The rules defined to refer the effects are not 100% accurate. GLCM_Correlation C9 <=8 >8

GLCM_InverseDifferenc
• The combinations were set according to the correlation and the eMoment C10 <=0.5 >0.5
calculation.
Future Implementation: NGTDM_Contrast C11 <=256 >256
NGTDM_Complexity C12 <=5 >5
• Need to check the pyradiomic related paper to find how to
segment / extract the features in medical image. NGTDM_Busyness C13 <=256 >256
NGTDM_Strength C14 <=1600 >1600
• Include the other features related to texture. NGTM_Coarseness C15 <=1600 >1600
• Partition the intensity with skewness.
CETA Cause-Effect Relationship
FOS_Mean (V1) and FOS_Skewness(V1) and FOS_Kurtosis (V1) and FOS_Entropy(V2) -> HIGH_Intensity
FOS_Mean (V2) and FOS_Skewness(V2) and FOS_Kurtosis (V2) and FOS_Entropy(V1) -> LOW_Intensity

NGTDM_Complexity(V2) and NGTDM_Coarseness(V1) and NGTDM_Busyness(V2) and NGTDM_Contrast(V1) ->


HIGH Complexity

NGTDM_Complexity(V1) and NGTDM_Coarseness(V1) and NGTDM_Busyness(V1) -> LOW_Complexity

GLCM_Entropy(V1) and GLCM_ASM(V2) and GLCM_InverseDifferenceMoment(V2) and GLCM_Correlation (V1)


and GLCM_Contrast(V2) -> HETEROGENEITY

GLCM_Entropy(V2) and GLCM_ASM(V1) and GLCM_InverseDifferenceMoment(V1) and GLCM_Correlation (V2)


and GLCM_Contrast(V1) -> HOMOGENEITY
Structural Similarity
• The structural similarity score can be found by reducing the high dimensional array to
2D array. [t-SNE (t-distributed stochastic neighbor embedding) algorithm]
• The t-SNE algorithm comprises two main stages. First, t-SNE constructs a probability
distribution over pairs of high-dimensional objects in such a way that similar objects are
assigned a higher probability while dissimilar points are assigned a lower probability.
Structural Similarity

• Tsne algorithm + kmeans


clustering ( cluster =10)
• Issue:
• According to kmeans, there are
two clusters.
• If you need similar image with
high probability, then we need to
increase the cluster points.

You might also like