Related Experiment Video
Updated: Aug 31, 2025
08:59
Morphology-Based Distinction Between Healthy and Pathological Cells Utilizing Fourier Transforms and Self-Organizing Maps
Published on: October 28, 2018
7.2K
Abnormality classification and localization using dual-branch whole-region-based CNN model with histopathological
Olaide N Oyelade1, Absalom E Ezugwu1, Hein S Venter2
1School of Mathematics, Statistics, and Computer Science, University of KwaZulu-Natal, King Edward Avenue, Pietermaritzburg Campus, Pietermaritzburg, 3201, KwaZulu-Natal, South Africa.
Computers in Biology and Medicine
|August 20, 2022
Summary
This study introduces a dual-branch deep learning framework for classifying and localizing abnormalities in medical images. The method achieves high accuracy without image annotation, aiding pathologists in abnormality detection.
Area of Science:
- Medical Imaging Analysis
- Computational Pathology
- Deep Learning Applications
Background:
- Medical image analysis, particularly abnormality classification and localization, is challenging.
- Deep learning shows promise but localizing abnormalities to support classification confidence remains an area of interest.
- Digital histopathology images present unique difficulties requiring advanced deep learning models.
Purpose of the Study:
- To develop a dual-branch deep learning framework combining classification and localization of abnormalities in medical images.
- To address the challenge of high-level deep learning models for digital histopathology.
- To automate pathology localization for improved acquisition planning and post-imaging analysis.
Main Methods:
- A dual-branch deep learning framework integrating Whole-image based CNN (WCNN) and region-based CNN (RCNN) architectures.
- Systematic combination of WCNN and RCNN for multi-class classification and localization without annotation-dependent images.
- Seamless confidence and explanation mechanism by mapping WCNN and RCNN outcomes.
Main Results:
- Achieved 97.08% classification accuracy and 94% localization accuracy on BACH and BreakHis datasets.
- Demonstrated simultaneous classification and localization of abnormalities.
- AUC for classification was 0.10.
Conclusions:
- A multi-neural network approach effectively addresses the combined problem of classification and localization of anomalies in digital medical images.
- The proposed method offers automated annotation of histopathology images.
- Provides support for human pathologists in accurately locating abnormalities.
