Deep Learning Outperforms Radiomics in Predicting Immunotherapy Outcomes in NSCLC

A new study reveals that deep learning significantly outperforms traditional radiomics in predicting how patients with metastatic non-small cell lung cancer (NSCLC) respond to immunotherapy.

Researchers analyzed data from three randomized trials-PEMBRO RT, NIVORAD, and MDACC-evaluating PD-1 inhibitor therapy, with or without stereotactic ablative body radiotherapy.

The deep learning model achieved an area under the curve (AUC) of 0.92 for predicting progressive disease per lesion, compared to just 0.57 for the radiomics-based model. This marks a substantial leap in predictive accuracy.

Radiomics-based survival models showed moderate performance, with concordance indices of 0.63 for overall survival and 0.59 for progression-free survival. Adding clinical variables like PD-L1 status and treatment arm improved these to 0.67 and 0.65, respectively.

Despite the promise, external validation on the NIVORAD and MDACC datasets showed reduced AUC values, highlighting challenges in model generalizability. The authors stress that further research is necessary before clinical adoption.