| Abstract Scope |
Immunotherapy efficacy depends on the immune landscape of tumors, yet current methods for evaluating immune infiltration rely heavily on invasive biopsies with limited sampling accuracy. We present SPARTICUS (Solves Pixel Analysis for Real-Time Treatment Insight using Characterization in Ultrasound Scans), a computational ultrasound-based approach designed to non-invasively assess tumor immune cell infiltration. Using tissue-mimicking hydrogels containing melanoma cells (B16F10), macrophages (RAW264.7), and mixed tumor–immune populations, we demonstrated that pixel intensity distributions from B-mode ultrasound images correlate with underlying cellular composition. In vivo validation in murine tumor models further revealed that ultrasound-derived pixel features predict CD45+ immune cell prevalence, with Lasso regression achieving an Rē of 0.848. Importantly, tumors treated with combined histotripsy and anti-SIRPα immunotherapy exhibited both the highest CD45 infiltration and distinct ultrasound intensity patterns, supporting the predictive capability of this method. A lightweight MobileNetV2-based segmentation model achieved a Dice coefficient of 90.56% for isolating tumor regions, forming the foundation for downstream analysis. Together, these results establish SPARTICUS as a low-cost, non-ionizing, real-time imaging strategy for immune response monitoring, with strong potential to reduce biopsy reliance and enable personalized immunotherapy guidance. |