Deep learning predicts chromosomal instability from histopathology images.

Area of Science:

• Oncology
• Computational Pathology
• Bioinformatics

Background:

• Chromosomal instability (CIN) is a key feature of human cancers but is difficult to test clinically.
• Hematoxylin and eosin (H&E) stained histology slides are widely available in clinical settings.

Purpose of the Study:

• To develop and validate a deep learning model for predicting CIN status from H&E histology images.
• To investigate the utility of histology-based CIN prediction in a large breast cancer cohort.

Main Methods:

• A deep learning model was trained on H&E images from The Cancer Genome Atlas (TCGA) breast cancer cohort.
• The model predicted CIN status (high vs. low) using histology data.
• Performance was evaluated using area under the curve (AUC), sensitivity, and specificity on an independent test set.

Main Results:

• The model achieved an AUC of 0.822, with 81.2% sensitivity and 68.7% specificity in classifying CIN status on the test set.
• Patch-level predictions revealed intra-tumor heterogeneity in CIN status within slides.
• High CIN scores in specific patches, rather than average scores, were more predictive of clinical outcome.

Conclusions:

• Deep learning models can accurately predict CIN status from routine H&E histology, offering a clinically accessible method.
• Intra-tumor heterogeneity in CIN is significant and has prognostic value, highlighting the importance of localized CIN assessment.