Abstract
Artificial intelligence (AI) is increasingly applied in oncology to enhance cancer detection, diagnosis, and treatment planning. Despite this progress, uncertainty remains regarding the robustness and generalizability of current AI applications in cancer imaging and pathology. This systematic review evaluated the evidence on AI applications in cancer imaging and pathology, synthesizing findings on their effectiveness, limitations, and implications for clinical practice. The review synthesized evidence from studies evaluating AI systems in cancer imaging and pathology, focusing on diagnostic performance, clinical utility, and methodological limitations. The review found that AI consistently demonstrated strong diagnostic performance across cancer types and imaging modalities, often matching or surpassing clinician accuracy. These systems showed particular promise in early cancer detection and decision support, with potential to reduce human error and support more personalized treatment strategies. However, limitations were evident: most studies lacked real-world clinical validation, integration with genomic and multimodal patient data was weak, and underrepresentation of minority groups raised concerns about generalizability and algorithmic bias. Moreover, issues of transparency, explainability, and ethical acceptability remain unresolved. The findings suggest that AI could function as an effective triage and decision-support tool in oncology, but safe and equitable implementation requires addressing current gaps in data diversity, validation, and clinical workflow integration. Future research should prioritize prospective studies in diverse populations and settings, ensuring that AI systems can be trusted and effectively embedded into routine cancer care for older adults.