Caris Life Sciences publishes new study on AI-based image analysis model

Caris Life Sciences (CAI) has published a new study in Communications Medicine, a Nature portfolio journal, demonstrating that Caris’ AI-based image analysis model has the potential to more accurately predict cancer biomarkers and patient survival than the conventional companion diagnostic, or CDx, methods. By analyzing hematoxylin and eosin, or H&E, images, the study demonstrated that Caris’ AI model can improve the assessment of critical cancer biomarkers and impact patient survival outcomes in breast and colorectal cancers. For this study, Caris’ AI model analyzed data from over 35,000 patients in the Caris clinico-genomic database. In breast cancer, the AI model scored PD-L1 positive phenotype status using an H&E image alone and assessed overall survival of patients treated with pembrolizumab, achieving a hazard ratio, or HR, for overall survival of 0.511, compared to an HR of 0.882 for traditionally scored PD-L1 IHCs, a result consistent with an almost doubling of overall survival for patients treated with pembrolizumab. In colorectal cancer, AI predicted mismatch repair deficiency, or MMRd, and microsatellite instability, or MSI, equivalent to traditional scoring.