Prostate Cancer Grade Assessment (PANDA)
Prostate Cancer Grade Assessment (PANDA): Automated Gleason Grading from Biopsy WSIs
This project is based on the Kaggle Prostate Cancer Grade Assessment (PANDA) dataset and competition. The dataset consists of whole-slide images (WSIs) of prostate biopsy samples stained with H&E, annotated by expert pathologists with Gleason grade groups. The goal is to develop machine learning models that automatically assess prostate cancer severity from digital pathology images.
https://www.kaggle.com/c/prostate-cancer-grade-assessment/data
Mentor Details:
Dr. Chen Sagiv
Mentor Details:
Requirments:
- Train deep learning models to predict Gleason grade groups from prostate biopsy whole-slide images.
- Develop patch-based or multi-instance learning strategies to handle gigapixel WSIs.
- Evaluate model performance using clinically relevant metrics such as Quadratic Weighted Kappa.
- Build reproducible training and inference pipelines suitable for research and benchmarking.
Problem Statement
Gleason grading is essential for prostate cancer diagnosis and treatment planning, but manual grading is time-consuming and subject to inter-observer variability. Whole-slide images are extremely large, making manual review and computational processing challenging. Automated AI-based approaches are needed to improve scalability, consistency, and reproducibility of prostate cancer grading.
Project Objectives
- Train deep learning models to predict Gleason grade groups from prostate biopsy whole-slide images.
- Develop patch-based or multi-instance learning strategies to handle gigapixel WSIs.
- Evaluate model performance using clinically relevant metrics such as Quadratic Weighted Kappa.
- Build reproducible training and inference pipelines suitable for research and benchmarking.
Technical Scope
- Handling and preprocessing of large whole-slide pathology images
- Patch extraction and color normalization
- Convolutional neural networks or transformer-based image models
- Aggregation of patch-level predictions to slide-level Gleason grades
- Model evaluation using accuracy and Quadratic Weighted Kappa metrics
Required Knowledge and Prerequisites
- Python programming
- Machine learning and deep learning fundamentals
- Experience with PyTorch or TensorFlow
Recommended Background:
- Digital pathology or medical image analysis
- Experience with large-scale image datasets
- Understanding of classification metrics and model validation
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 AI model capable of predicting Gleason grade groups from biopsy WSIs
- Quantitative benchmark results evaluated with Quadratic Weighted Kappa
- A documented and reproducible data processing and modeling pipeline
- Insights into challenges of large-scale computational pathology