Deep Learning Models for Poorly Differentiated Colorectal Adenocarcinoma Classification in Whole Slide Images Using Transfer Learning
Problem
Identifying poorly differentiated colorectal adenocarcinoma in endoscopic biopsy whole slide images is difficult and could benefit from AI tools for pathologists.
Why it matters
Poorly differentiated colorectal ADC carries a poor prognosis yet occurs in under 5% of cases, making screening detection hard.
What the AI did
Deep learning models trained via a simple transfer learning method classify poorly differentiated colorectal ADC from whole slide images, evaluated on test sets from five distinct sources.
Results
ROC AUCs up to 0.95 on 1799 test cases from five distinct sources.
Impact
Evidence supports strong classification performance as a pathologist aid, not clinical deployment.
Record
| Venue | Diagnostics |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Masayuki Tsuneki, Fahdi Kanavati |
| Citations | 28 (OpenAlex, at import time) |
| References | 52 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/diagnostics11112074 |
| Open access | gold |
| Keywords | colorectal adenocarcinoma, transfer learning, whole slide images, classification, AUC, digital pathology |
Importance rating
real but narrower advance
Abstract states poorly differentiated ADC has poor prognosis but occurs in under 5% of colorectal carcinomas; AI pathologist aid for screening is a real but narrower advance.