A research update highlights a label-free microfluidic platform combined with YOLOv8 deep learning to enrich and identify rare circulating tumor cells from blood. The system aims to improve sensitivity in detection workflows that are limited by low event counts and labeling requirements. Separately, a new approach to classify cancer subtypes uses a graph neural network (CALΤ-GNN) designed for imbalanced long-tail molecular datasets derived from TCGA, targeting subtype classification where traditional training struggles. Together, the developments point to two converging needs in oncology tools: handling sparse biology in the lab and improving model performance under real-world distribution shifts in computational pipelines.