A deep-learning framework to predict cancer treatment response from histopathology images through imputed
Danh-Tai Hoang1, Gal Dinstag2, Eldad D Shulman3
1Biological Data Science Institute, College of Science, Australian National University, Canberra, Australian Capital Territory, Australia. danhtai.hoang@anu.edu.au.
Nature Cancer
|July 3, 2024
Summary
Artificial intelligence (AI) can now predict cancer treatment response using standard tumor slides. ENLIGHT-DeepPT, an AI tool, infers gene expression from slides to forecast patient response to targeted and immune therapies.
Area of Science:
- Computational biology
- Digital pathology
- Precision oncology
Background:
- Hematoxylin and eosin-stained tumor slides are routinely used in cancer diagnosis.
- Artificial intelligence (AI) offers new avenues for extracting complex biological information from histopathology images.
Purpose of the Study:
- To develop and validate an AI-driven framework (ENLIGHT-DeepPT) for predicting patient response to cancer therapies using digital pathology.
- To assess the performance of ENLIGHT-DeepPT in diverse cancer types and treatment settings.
Main Methods:
- ENLIGHT-DeepPT is a two-step approach: DeepPT predicts genome-wide mRNA expression from slides, and ENLIGHT predicts therapy response from inferred expression.
- The framework was validated across 16 The Cancer Genome Atlas (TCGA) cohorts and five independent patient cohorts.
Main Results:
- DeepPT accurately predicted transcriptomics in all tested TCGA cohorts and generalized to independent datasets.
- ENLIGHT-DeepPT successfully identified true responders across six cancer types and four treatments, demonstrating a 39.5% increased response rate.
- Prediction accuracy was achieved without direct training on treatment data, rivaling methods that require such specific training.
Conclusions:
- ENLIGHT-DeepPT provides a powerful, non-invasive method for predicting cancer therapy response from standard histology slides.
- This AI framework holds significant potential for advancing precision oncology by guiding treatment decisions.
