Machine learning continues to move from retrospective performance toward tools that can influence patient pathways. A multicenter study reports that CT-based radiomics can noninvasively predict microsatellite instability in gastric cancer with clinically useful accuracy, potentially reducing the need for immediate biopsy solely to determine dMMR/MSI status. In breast cancer, another study claims MRI-based machine learning distinguishes luminal A from luminal B subtypes without biopsy by combining features from three MRI sequences. While still in diagnostic validation territory, the work aims to streamline subtype decision-making earlier in care. Across the set, the recurring theme is patient selection: using imaging-derived biomarkers to make treatment decisions faster while maintaining sufficient diagnostic agreement for clinical action.