講演情報

[E4-4]Deep learning in radiogenomics for enhanced risk prediction only from CT images in colorectal cancer

Feng Gao1,2,3, Fengao Wang4,5, Chuling Hu1,2,3, Du Cai1,2,3, Yibin Zhao6, Daisuke Izumi7, Haoning Qi1,2,3, Baowen Gai1,2,3, Junxiang Ding4,5, Ruikun He8, Junwei Liu5, Yixue Li4,5,9,10,11,12,13, Xiaojian Wu1,2,3 (1.Department of General Surgery (Department of Colorectal Surgery), The Sixth Affiliated Hospital, Sun Yat–sen University, 2.Guangdong Provincial Key Laboratory of Colorectal and Pelvic Floor Diseases, The Sixth Affiliated Hospital, Sun Yat–sen University, 3.Biomedical Innovation Center, The Sixth Affiliated Hospital, Sun Yat–sen University, 4.Institute for Advanced Study, University of Chinese Academy of Sciences, 5.Guangzhou National Laboratory, 6.Department of Colorectal Surgery, Ningbo Medical Center Lihuili Hospital (Affiliated Lihuili Hospital of Ningbo University), 7.Izumi Gastroenterology & Surgery Clinic, 8.BYHEALTH Institute of Nutrition & Health, 9.GZMU–GIBH Joint School of Life Sciences, The Guangdong–Hong Kong–Macau Joint Laboratory for Cell Fate Regulation and Diseases, Guangzhou Medical University, 10.School of Life Sciences and Biotechnology, Shanghai Jiao Tong University, 11.Shanghai Institute of Nutrition and Health, Chinese Academy of Sciences, 12.Collaborative Innovation Center for Genetics and Development, Fudan University, 13.Shanghai Institute for Biomedical and Pharmaceutical Technologies)
PDFダウンロードPDFダウンロード
Background: Accurate prognosis prediction in colorectal cancer (CRC) patients is clinically essential. While the efficiency of radio-genomics multimodal learning in prognosis prediction, its clinical implementation is high costs and difficult. We aimed to develop a deep learning model to integrate radio-genomics datasets and enable prognosis prediction using only CT images.
Methods: Our retrospective study involved two CRC cohorts from the Sixth Affiliated Hospital of Sun Yat-sen University, who had paired radio-genomic data (n=486) or only CT images (n=3004). We developed a Cross-Infer Survival Multimodal (CISM) model that predicts overall survival in CRC patients trained with radio-genomic data and is capable of prognosis prediction with only CT images. We evaluated the performance improvement of our model in prognosis prediction with only CT images and characterized the important multi-omics features in patient survival.
Results: With the prospective training cohort consisting of paired CT images and genomic data, the CISM model can predict the overall survival of CRC patients with multimodal inputs (C-index 0.701), only CT images input (C-index 0.658), and surpassing the CT image model (C-index 0.619). In the validation cohort with only CT images, the CISM model demonstrated higher performance in stratifying CRC patients into high-risk and low-risk groups (HR 2.06) compared to CT image model (HR 1.37). We explored the genomic and CT image features related to the prognosis of CRC patients and found the optimal image lesion focuses with the CISM model.
Conclusions: The CISM model shows superior performance in prognosis prediction with only CT images, suggesting that cross-modal interactions benefit clinical decision-making with limited clinical resources.