▌ Project lead

    Arvydas Laurinavicius ResearchGate 

▌ Project team

     Renaldas Augulis ResearchGate 

    Allan Rasmusson ResearchGate 

    Yu-Chieh Lin ResearchGate 

▌ Project overview

This activity is focused on advanced digital pathology data management and machine learning methods to for predictive modeling of disease. This includes on spatially informed pathology image analysis, such as hexagonal grid–based spatial analytics, principal component analysis (PCA), clustering, and graph-based analysis of tissue architecture. Federated learning set up was tested for multicenter collaborative model development without direct data sharing, along with survival analysis integrating clinical and pathology imaging-derived metrics.

The work started with establishment of technical infrastructure and the research team. Specialized digital pathology software (HALO) and a GPU cluster–based analytical environment were deployed, and two researchers were recruited. Following bioethics approvals, four large-scale, fully anonymized biomedical image datasets were constructed, including more than 7,300 whole slide images (WSIs). The datasets cover renal cell carcinoma, breast tissue, native and transplanted kidney biopsies, and non–muscle-invasive urothelial carcinoma of the bladder.

A federated learning–based artificial intelligence model was developed to predict recurrence of early-stage urothelial carcinoma using data from 2 medical centers in Lithuania and Taiwan (under review). In parallel, two advanced analytical methodologies are developed to characterize kidney microarchitecture through spatial and graph-based modeling.

▌ Impact

The project contributes to the development of reliable, interpretable, and clinically applicable artificial intelligence solutions in pathology. It strengthens international and multidisciplinary collaboration and opens further collaborations in medical data science in development and validation of AI-based tools.

▌ Public Dissemination

The project results were presented at the joint conference of the Lithuanian Artificial Intelligence Association and the Lithuanian Academy of Sciences, “Artificial Intelligence Technologies in Medicine: Research and Diagnostics,” on October 26, 2024. A multicenter AI model for the diagnosis and prognosis of early-stage urothelial carcinoma was publicly introduced, including explainable heatmaps that support transparent interpretation of model decisions. The presentation reached a broad audience of AI, medical, and oncology researchers and professionals.