Related Experiment Video
Updated: May 29, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Automating Prostate Cancer Grading: A Novel Deep Learning Framework for Automatic Prostate Cancer Grade Assessment
Saidul Kabir1, Rusab Sarmun1, Rafif Mahmood Al Saady2
1Department of Electrical and Electronic Engineering, University of Dhaka, Dhaka, 1000, Bangladesh.
Journal of Imaging Informatics in Medicine
|February 6, 2025
Summary
This study introduces an automated deep learning system to improve prostate cancer (PCa) grading accuracy. The innovative framework enhances diagnostic precision, offering a reliable tool for assessing PCa severity and aiding clinical decisions.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Prostate cancer (PCa) grading relies on the subjective Gleason system, necessitating more objective methods.
- Current diagnostic variability highlights the need for automated, reliable tools to improve accuracy and minimize human error.
- Deep learning offers potential for enhanced precision in medical image analysis and cancer grading.
Purpose of the Study:
- To develop and evaluate an innovative three-stage deep learning framework for automated prostate cancer grading.
- To improve the accuracy and reliability of PCa severity assessment using automated systems.
- To provide a prospective prognostic tool for clinically significant and efficient PCa evaluation.
Main Methods:
- A three-stage deep learning framework was developed using the PANDA challenge dataset (2699 cases).
- Stages included: classification of PCa grades (benign, Gleason 3-5) using DNNs, segmentation of PCa grades, and ISUP grade computation via ML classifiers.
- Optimized patch sampling, DeepLabV3 with Self-ONN, and EfficientNet_b0 were employed for classification and segmentation.
Main Results:
- EfficientNet_b0 achieved an 83.83% F1-score for classification.
- DeepLabV3+ with Self-ONN and EfficientNet encoder yielded an 84.9% Dice Similarity Coefficient (DSC) for segmentation.
- The framework achieved a 0.9215 quadratic weighted kappa (QWK) score using a RandomForest classifier for final ISUP grade prediction.
Conclusions:
- The proposed deep learning framework demonstrates promising results for automated prostate cancer grading.
- The system offers a reliable and efficient approach for assessing PCa severity, potentially serving as a prognostic tool.
- Further research is needed to validate the framework's adaptability across diverse clinical settings.
