Comprehensive Summary
Continuing to improve upon classification models and risk stratification is a key aspect of artificial intelligence (AI) integration. Du et al utilized a deep learning model to evaluate patients undergoing a partial nephrectomy, incorporating CT radiomics and clinical features. Surgical approach and warm ischemia time (WIT) were excluded from the model to prevent bias and focus on preoperative complication prediction. Complication prediction and risk grading for patients with low and no risk, this model proved to be norinferior in performance to current anatomical classification models, RENAL and PADUA.
Outcomes and Implications
This deep learning model can provide more comprehensive patient outcome predictions through the analysis of radiomic features such as texture, shape, and intensity. Grayscale and texture are other components more easily differentiated by this model when compared to classic anatomical models. Furthermore, they attest to the superiority in incorporating bleeding complications based on cancer types, such as in cystic versus solid renal cancer. While AI may have biases of its own, this model also works to eliminate biases and limitations present in expert-based models.