Deep learning–based clinical decision support for renal cell carcinoma using computed tomography: an expert-informed scoping review

Aims Deep learning (DL) has the potential to substantially support clinical decision making in patients with renal cell carcinoma (RCC), especially regarding computed tomography (CT). The aim of this expert-informed scoping review is to place CT-based DL research within the context of clinical practice, supported by a clinician-driven survey assessing clinical relevance and workload. We introduce a structure based on the role of DL applications in clinical practice: Automation, Augmentation, and Innovation (AtAgIn). Methods We conducted a systematic review of the literature. Furthermore, we surveyed clinicians to identify the clinical need for DL-driven tasks, and articles without sufficient clinical relevance or workload were excluded during full text screening. Results: We included 80 articles focused on clinically driven DL tasks in RCC patients. A survey among nine radiologists and five urologists revealed a high clinical interest across AtAgIn categories. Workload scores were more variable, with higher workloads for tasks requiring manual segmentation and anatomical quantification. Most included articles fell within the innovation category, especially classification of tumor type and malignancy and prediction of outcome and survival. In contrast, tasks related to disease-specific information, especially in lymph node status and presence of distant metastases, were limited. Moreover, literature about surgical planning, per- and post-operative outcomes and general health status of patients was scarce. Conclusions: This expert informed scoping review shows that CT-based DL research in renal cell carcinoma is unevenly aligned with clinical needs. Multi-center datasets, standardizes endpoints and specific validation strategies, aligned with the intended clinical function, are needed.