This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
This paper highlights the importance of accurately predicting gene mutations, mutation subtypes, and exons in lung cancer patients for personalized treatment planning and prognostic assessment. To address regional imbalances in medical resources and the high cost of genome sequencing, we propose a method for inferring these mutations and exon alterations using artificial intelligence from routine histopathology images. We constructed the PathGene dataset, combining histopathology images and next-generation sequencing reports from 1,576 patients at the Second Hyangya Hospital of Chungnam University and 448 patients from the TCGA-LUAD study. We then linked whole-slide images to key gene mutation status, mutation subtypes, exons, and tumor mutation burden (TMB) status. Unlike existing datasets, PathGene provides molecular-level information related to histopathology images, facilitating the development of biomarker prediction models. We applied 11 multi-instance learning methods to PathGene to benchmark mutation, subtype, exon, and TMB prediction tasks. This support early genetic screening and the development of personalized, targeted treatment plans for lung cancer patients. Code and data are available at https://github.com/panliangrui/NIPS2025/ .
Presenting a novel AI-based method that could contribute to early genetic screening for lung cancer and the development of personalized precision treatment plans.
◦
The PathGene dataset provides a useful resource for developing biomarker prediction models by combining histopathological images with molecular-level information.
◦
Selecting the optimal model and suggesting future research directions through comparative analysis of the performance of various multi-instance learning methods.
•
Limitations:
◦
The PathGene dataset is limited to specific hospitals and TCGA data, requiring further research on generalizability.
◦
A comparative study of the performance of other approaches other than the multi-instance learning method used in this study is needed.
◦
Additional research and clinical trials are needed to verify clinical utility.
