VU and Taiwan Researchers Develop Model to Identify Treatment-Resistant Bladder Cancer

A team of researchers from Vilnius University (VU), led by Prof. Arvydas Laurinavičius, together with colleagues from the Urology Centre at VU Hospital Santaros Klinikos and the National Centre of Pathology, as well as partners from Taiwan, has published the findings of a study on urothelial carcinoma prognosis in the prestigious journal, npj Precision Oncology, part of the Nature Portfolio Group. The study was carried out in collaboration among urologists, pathologists, data scientists, and artificial intelligence (AI) researchers from Lithuania and Taiwan. The co-operation between specialists from different fields involved integrating clinical, pathological, and digital pathology data for the development of predictive models.

Non-muscle-invasive papillary urothelial carcinoma (NMIPUC) is the most common type of bladder cancer. The standard treatment for this disease is transurethral resection of the tumour followed by long-term adjuvant intravesical Bacille Calmette–Guérin (BCG) immunotherapy, which significantly reduces the risk of recurrence and disease progression. However, in approximately 40% of patients, tumours are resistant to this form of immunotherapy and progress despite the administered treatment. It is therefore crucial to accurately predict the course of the disease and identify patients likely to be resistant to treatment, who will require alternative strategies.

According to Prof. Laurinavičius, as the number of NMIPUC treatment methods for bladder cancer increases, it is vital to predict the course of the disease when making clinical decisions: “This is an area of intensive research. In addition, the models we have developed reveal just how much valuable information is contained in the microscopic image of a tumour.”

An AI model has been developed to predict the course of the disease. It has been trained using a federated learning (FL)-based AI framework, which enables us to create shared predictive algorithms from data accumulated across different institutions, without transferring it to a single central database. This model improves the predictive accuracy for tumour staging and grading in digital histopathology images, and identifies novel histological risk factors derived from these images, which allows us to predict patients’ response to treatment, and, based on the characteristics of tumour microarchitecture, allows us to stratify patient groups according to different relapse prediction risk profiles.

Collaboration with Taiwanese researchers began in 2023, when Dr Yu-Chieh Lin attended the international life sciences forum, Life Sciences Baltics, in Lithuania. “He got thoroughly involved with the work of our digital pathology team, and we established a collaboration with our Taiwanese colleagues. What makes this work particularly interesting is that we operate within our own digital environments, where the algorithms “travel and learn”. Further results have been achieved, and there are no plans to stop,” Laurinavičius commented.

The project is funded by the Research Council of Lithuania under the Ministry of Education, Science and Sport’s programme, University Excellence Initiatives (the ministry’s Science Development Programme progress measure No. 12-001-01-01-01 “Improving the Environment for Science and Studies”). Project contract No. S-A-UEI-23-11.