Work

Evaluating the Performance of LYDIA: An AI-Powered Assistant in the Detection of Metastatic Tumors to Optimize Clinical Workflows and Inform Soft Tissue Surgical Decision-Making

Year: 2026

Type: article

Abstract: The integration of artificial intelligence (AI) into digital histopathology has the potential to improve the accuracy and efficiency of metastatic cancer diagnosis. We evaluated LYDIA (LYmph noDe assIstAnt), an AI-based decision-support system, for both standalone diagnostic performance and its impact on histopathologists' workflow, diagnostic accuracy, and resource utilization. LYDIA was evaluated on a blinded dataset of 366 whole-slide images (WSIs) from breast, colorectal, lung, and skin cancers. Standalone performance demonstrated excellent discrimination, achieving ROC-AUC values of 0.995, 0.963, 0.973, and 0.983 for breast, colorectal, lung, and skin cancers, respectively. Clinical utility was further assessed in a multi-reader study involving four experienced histopathologists interpreting 105 WSIs with and without AI assistance. AI-assisted diagnosis significantly reduced time-to-diagnosis across all metastasis sizes, with a maximum 1.59-fold acceleration for micro-metastases, corresponding to a mean time saving of 26.5 s per WSI. LYDIA also improved diagnostic sensitivity from 77.3% to 87.3%. These findings demonstrate that LYDIA can enhance both the efficiency and accuracy of lymph node metastasis detection while reducing diagnostic workload and the need for ancillary testing. Beyond improving routine pathology workflows, the system's rapid inference capabilities and human-expert level performance may support future intraoperative diagnostic applications, enabling timely and automatic or semi-automatic assessment of nodal status to inform surgical decision-making. The resulting reductions in diagnostic time and ancillary testing costs have the potential to improve healthcare resource utilization and patient care, mainly in soft tissue reconstruction and surgical repair decisions.

Source: Bioengineering

DOI: 10.3390/bioengineering13091021