Cancer immunotherapy response prediction from multi-modal clinical and image data using semi-supervised deep...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Immunotherapy is a key cancer treatment, but predicting patient response remains challenging due to a lack of reliable biomarkers.
• Deep learning has advanced cancer detection but has had limited success in predicting treatment response.

Purpose of the Study:

• To predict immunotherapy response in gastric cancer patients using routinely available clinical and computed tomography (CT) imaging data.
• To develop a multi-modal deep learning radiomics approach for enhanced prediction accuracy.

Main Methods:

• A multi-modal deep learning radiomics model was developed using clinical data and CT images from 168 advanced gastric cancer patients treated with immunotherapy.
• A semi-supervised framework utilized an additional dataset of 2,029 patients (not on immunotherapy) to learn intrinsic imaging phenotypes, addressing small training data limitations.
• Model performance was validated in two independent cohorts comprising 81 immunotherapy-treated patients.

Main Results:

• The deep learning model achieved an area under the receiver operating characteristics curve (AUC) of 0.791 in the internal validation cohort and 0.812 in the external validation cohort.
• Integrating PD-L1 expression data with the model further improved prediction accuracy by 4-7% in absolute terms.

Conclusions:

• The deep learning model demonstrates promising performance in predicting immunotherapy response using standard clinical and imaging data.
• The multi-modal approach is adaptable and can be enhanced by incorporating additional relevant data for improved immunotherapy response prediction.