Related Experiment Video
Updated: Aug 5, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Self-supervised attention-based deep learning for pan-cancer mutation prediction from histopathology.
Oliver Lester Saldanha1,2, Chiara M L Loeffler1,2, Jan Moritz Niehues1,2
1Else Kroener Fresenius Center for Digital Health, Medical Faculty Carl Gustav Carus, Technical University Dresden, Dresden, Germany.
NPJ Precision Oncology
|March 28, 2023
Summary
Deep learning accurately predicts genetic alterations from tumor histology slides. This method shows strong generalizability across different datasets, improving diagnostic capabilities.
Area of Science:
- Computational pathology
- Genomics
- Artificial intelligence in medicine
Background:
- Tumor histopathology reflects underlying genetic alterations.
- Deep learning models show promise in predicting genetic mutations from pathology images.
- Generalizability of these deep learning models to external datasets remains a challenge.
Purpose of the Study:
- To systematically evaluate the predictability and generalizability of deep learning models for predicting genetic alterations from histology.
- To assess the performance of an integrated analysis pipeline combining self-supervised feature extraction and attention-based multiple instance learning.
Main Methods:
- Utilized two large, multi-tumor type datasets for comprehensive analysis.
- Employed a deep learning pipeline integrating self-supervised feature extraction.
- Incorporated attention-based multiple instance learning for robust prediction.
Main Results:
- The proposed analysis pipeline demonstrated robust predictability of genetic alterations from histology slides.
- The model exhibited strong generalizability across independent, external datasets.
- Validated the effectiveness of the integrated self-supervised and attention-based approach.
Conclusions:
- Deep learning, particularly with the presented pipeline, offers a reliable method for predicting genetic alterations from histopathology.
- The findings support the potential of computational pathology for non-invasively inferring tumor genetics.
- This approach enhances the utility of digital pathology for precision oncology.
