STAT390 CMIL Classification
Aim: The objective of this project is to identify the severity of potential eye cancer by looking at a particular eye tissue of the patient.
In medical terms, we want to develop a machine learning model to accurately classify Conjunctival melanocytic intraepithelial lesions (C-MIL) as per the WHO 2022 classification system. Providing a reproducible and accurate grading of C-MIL will help doctors select the most appropriate management plan for the patient.
The Northwestern University STAT390 Class has made the following progress on this project:
Step 1: Extracting tissue slices from Whole Slice Images (WSI) using QuPath
Step 2: Matching similar tissue slices across the different stains (H&E, Melan-A, Sox-10) for each patient when there is a correct match
Step 3: Send matched slices to pathologist team, who will annotate H&E slice with high grade, low grade, and benign regions and send back to us
Step 4: While the pathologist team annotates slices, orienting matched slices and apply patching across epithelium for matched slices. Currently looking at 3 different approaches to patching to find the optimal results
Step 5: Create a machine learning model to classify slices as low-grade, high-grade, or benign
Literature Review: Researched adaptive pooling, cross view transformer, square patches (advantages, sizes, and rationale), padding effect, etc.