Predicting tumour mutations from pathology images may help increase personalised medicine utilisation

Share :
Published: 17 Oct 2024
Views: 19
Rating:
Save
Dr Rom Leidner - Providence Cancer Institute, Portland, USA

Dr Rom Leidner speaks to ecancer about an application of GigaPath: An open-weight billion-parameter AI foundation model based on a novel vision transformer architecture for cancer mutation prediction and TME analysis.

GigaPath is an open-weight billion-parameter AI foundation model that has been pre-trained on an extensive digital pathology dataset sourced from 28 cancer centres. This dataset comprises 1,384,860,229 image tiles extracted from 171,189 H&E slides of biopsies and resections from over 30,000 patients, encompassing 31 primary tissue types.

In this study, GigaPath's H&E molecular predictions were compared with competing methods such as HIPT, CtransPath, and REMEDIS across three distinct tasks: predicting lung adenocarcinoma using a 5-gene panel (EGFR, FAT1, KRAS, TP53, LRP1B), the pan-cancer 5-gene panel, and estimating tumour mutation burden.

The results show that GigaPath can potentially be applied to broader biomedical domains for efficient self-supervised learning from high-resolution images and help increase personalised medicine utilisation.