Abstract
Schistosomiasis, a neglected tropical disease affecting millions globally, urgently requires new therapies. Current treatments, like praziquantel (PZQ), face challenges such as drug resistance and ineffectiveness against juvenile parasites. This study develops an automated, high-throughput computational platform to quantify schistosome viability from video data. We introduced an end-to-end system that leverages foundation models in computer vision and fine-tuning to assess schistosome viability from videos. By fine-tuning advanced image segmentation and spatiotemporal feature representation models, our approach accurately captures both morphological and motility-related features of schistosome and maps them to worm viability directly. As a proof of concept, we constructed two datasets (a PZQ-treatment video dataset with 325 videos and a multi-compound treatment video dataset with 245 videos), designed three worm viability assessment tasks and performed extensive evaluation on them. In addition, we developed a schistosome viability scoring tool, which can be accessed online. The system achieved superior predictive accuracy in PZQ-treated worms, with a Pearson correlation coefficient (PCC) of 0.937 for concentration regression, outperforming approaches like hand-crafted feature methods and wrmXpress. A novel 24-hour equivalent PZQ concentration metric was introduced, addressing saturation effects and showing strong generalizability across 13 other compounds (PCC = 0.712). Direct viability score predictions correlated highly with expert assessments, with PCCs of 0.892 for multi-worm analyses and 0.831 for individual worms. We developed a scalable, automated platform for anti-schistosomal drug discovery, providing reliable viability assessments. An accessible online tool enables efficient screening and has broader implications for parasitological research.