The use of artificial intelligence (AI) into clinical decision support systems (CDSSs) is a promising advancement in precision oncology. By analysing vast and complex datasets from multiple modalities, including radiology, histology, genomics, and electronic health records, AI can help clinicians make more comprehensive and personalized treatment recommendations. This approach moves beyond traditional biomarker-based patient stratification to uncover new patterns and associations within and across diverse data types. AI applications in this field encompass various tasks, such as risk assessment, improving diagnostic accuracy, and predicting patient outcomes, including survival, treatment response, and recurrence. An essential benefit of AI-driven Clinical Decision Support Systems (CDSS) is their capacity to synthesise information that exceeds the complexity of subjective human interpretation, thus facilitating the identification of novel and specific biomarkers. AI models have demonstrated high consistency with evaluations like diagnosing pulmonary nodules from CT scans. However, significant challenges remain, including the need for robust data security, effective data representation, and, critically, the explainability of AI-based predictions. This review generates evidence from almost 139 peer-reviewed studies published upto 2025, encompassing a huge patient record across solid and hematological malignancies. Compared with traditional biomarker-based stratification, multimodal AI models demonstrate superior predictive performance, with reported AUC values typically ranging from 0.82 to 0.96, compared with 0.65–0.78 for single-modality or rule-based clinical models. The role of explainability in CDSS is a debated topic, with its added value dependent on factors such as technical feasibility, validation levels, implementation context, and the user group. Addressing these challenges will be vital for the realistic and widespread adoption of AI in multidisciplinary cancer treatment and for achieving ground-breaking progress in personalized medicine.
Borat et al. (2026) studied this question.