Comprehensive Summary
Rakaee et al focuses on how an artificial intelligence pathology model can identify non-small cell lung cancer (NSCLC) patients who would benefit from immune checkpoint inhibitor treatment (ICI). A deep learning-based stratification model, DEEPIO, was developed to predict efficiency of ICI from digital hematoxylin-eosin stains of pathology specimens from NSCLC patients. DEEPIO was developed and validated with patient datasets from Dana-Farber Cancer Institute and 3 centers in the European Union. DEEPIO assigns each image a score from 0 to 1, indicating likelihood of response to ICI. It correctly predicted nonresponder status status for 55% of patients and responder status for 15%, with an overall accuracy of 70%. The DEEPIO score was significantly associated with survival with firstline or subsequent ICI. DEEPIO, combined with a predictive biomarker, programmed death ligand-1, also had a higher positive predictive value (0.42) and negative predictive value (0.86) compared to individual biomarkers such as tumor mutational burden. Across the datasets, DEEPIO shows great promise in predicting clinical outcomes of ICL therapy and surpassed established biomarkers.
Outcomes and Implications
This is the first study that explored how an artificial-response model can predict ICI therapy with advanced stages of NSCLC using digital H&E pathology scans. Seeing how DEEPIO did better than our established biomarkers shows potential for the integration of AI into clinical diagnosis and choice for treatment for NSCLC patients. The model focuses on tumor epithelial and inflammatory reaction subregions, suggesting that the model might be using immunological features to make its predictions. Future testing should be done to test DEEPIO on other datasets from other cancer centers in the United States and the world to improve its predictive accuracy. DEEPIO fell short of the ideal threshold of 0.8, so further changes such as vision transformers are needed to the model. DEEPIO also has the potential of being expanded to adjuvant or neoadjuvant ICI or chemo-ICI therapies in early-stage NSCLC.