Gleason 2019
Gleason 2019: Automated Prostate Cancer Gleason Grading
This project is based on the Gleason 2019 Grand Challenge, a benchmark challenge introduced at MICCAI 2019 focused on automated Gleason grading of prostate cancer from H&E-stained digital pathology images. The challenge evaluates algorithms that predict Gleason patterns at pixel level and Gleason scores at tissue core level using expert-annotated data.
https://gleason2019.grand-challenge.org/
Mentor Details:
Dr. Chen Sagiv
Mentor Details:
Requirments:
- Develop machine learning models for pixel-level Gleason pattern classification.
- Predict core-level Gleason scores from tissue micro-array images.
- Benchmark algorithm performance using standardized evaluation metrics.
- Analyze agreement between automated predictions and expert annotations.
Problem Statement
Gleason grading is a cornerstone of prostate cancer diagnosis and prognosis, but manual grading is time-consuming and subject to inter- and intra-observer variability among pathologists. Automated and reproducible AI-based approaches are needed to support clinical workflows and improve grading consistency.
Project Objectives
- Develop machine learning models for pixel-level Gleason pattern classification.
- Predict core-level Gleason scores from tissue micro-array images.
- Benchmark algorithm performance using standardized evaluation metrics.
- Analyze agreement between automated predictions and expert annotations.
Technical Scope
- Processing of H&E-stained prostate cancer tissue images
- Image preprocessing and patch-based modeling
- Deep learning for semantic segmentation and classification
- Aggregation of pixel-level predictions to core-level Gleason scores
- Model evaluation using accuracy, kappa statistics, and challenge metrics
Required Knowledge and Prerequisites
Required Knowledge and Prerequisites
Core Requirements:
- Python programming
- Machine learning fundamentals
- Deep learning frameworks (PyTorch or TensorFlow)
Recommended Background:
- Digital pathology and histopathology imaging
- Semantic segmentation models (e.g., U-Net)
- Medical image evaluation metrics
Project Difficulty and Expected Level
Overall Difficulty: varies
This project is well-suited for:
Teams of 2–4 students
This project can also be done coding free with the DeePathology STUDIO.
Expected Outcomes
- A trained Gleason grading model for prostate cancer tissue
- Quantitative benchmark results on validation and test sets
- A reproducible evaluation pipeline
- Improved understanding of challenges in computational pathology