Deep Morphology Learning Enhances Ex Vivo Drug Profiling-Based Precision Medicine
Drug testing in patient biopsy-derived cells can identify potent treatments for patients suffering from relapsed or refractory hematologic cancers. Here we investigate the use of weakly supervised deep learning on cell morphologies (DML) to comple-ment diagnostic marker-based identifi cation of mali...
Main Authors: | , , , , , , , , , , , , , , |
---|---|
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
American Association for Cancer Research
2022
|
Subjects: | |
Online Access: | https://hdl.handle.net/20.500.11850/584166 https://doi.org/10.3929/ethz-b-000584166 |
Summary: | Drug testing in patient biopsy-derived cells can identify potent treatments for patients suffering from relapsed or refractory hematologic cancers. Here we investigate the use of weakly supervised deep learning on cell morphologies (DML) to comple-ment diagnostic marker-based identifi cation of malignant and nonmalignant cells in drug testing. Across 390 biopsies from 289 patients with diverse blood cancers, DML-based drug responses show improved reproducibility and clustering of drugs with the same mode of action. DML does so by adapting to batch effects and by autonomously recognizing disease-associated cell morphologies. In a post hoc analysis of 66 patients, DML-recommended treatments led to improved progression-free survival compared with marker-based recommendations and physician's choice-based treatments. Treatments recommended by both immunofl uorescence and DML doubled the fraction of patients achieving exceptional clinical responses. Thus, DML-enhanced ex vivo drug screening is a promising tool in the identifi cation of effective personalized treatments.SIGNIFICANCE: We have recently demonstrated that image-based drug screening in patient samples identifi es effective treatment options for patients with advanced blood cancers. Here we show that using deep learning to identify malignant and nonmalignant cells by morphology improves such screens. The presented workfl ow is robust, automatable, and compatible with clinical routine. ISSN:2643-3249 ISSN:2643-3230 |
---|