Tjalling Bosse

LUMC

In short, Tjalling Bosse's research entails the following:

Deep uLMS: Deep Learning To Improve Uterine LeiomyoSarcoma Diagnostics

Optimal treatment for patients with uterine leiomyosarcoma (uLMS) starts with an accurate diagnosis, which should be reproducible, prognostic (i.e. provide information on tumor behavior) and predictive (i.e. provide information as to the benefit of certain treatments). Accurate diagnosis is also critically important to correctly assign patients into uLMS-specific clinical trials. The diagnosis of uLMS is provided by a physician with specialty training in pathology who analyzes the tumor tissue with a microscope. Microscopic features, such as shape and size of the tumor cells (“nuclear atypia”), rate of proliferation (“mitotic activity”) and cell death (“tumor necrosis”) are important factors that separate uLMS from its benign counterpart ("fibroids" or "leiomyomas"). This can be a challenging task, as benign fibroids can mimic the appearance of a uLMS. In addition to rendering a diagnosis, the pathologist is also tasked to provide tumor characteristics that are predictive of tumor behavior. In non-uterine LMS a robust and broadly applied grading system has been established to distinguish LMS with low risk ("low grade") from those with high risk of recurrence and/or distant spread ("high grade").

Despite several attempts, such a grading system has not been fully established for uLMS patients. A robust risk-stratification system would be particularly valuable for uLMS patients who present with a tumor confined to the uterus (low stage, majority of patients), who are treated surgically (removal of the uterus) with the intention for complete cure. These patients would benefit from a risk-stratification system that allows prediction of their chances of recurrent disease. We hypothesize that predicting the outcome of patients with uLMS following surgery by certain microscopic features might be feasible when using state-of-the-art artificial intelligence (AI) techniques. We formed a team of experts with experience in this field to execute this project.

In this project we will build a platform for characterizing uLMS, based on advanced AI-techniques, by combining genetic and histology markers. More specifically, we will investigate the relationship between genetic markers, histology and disease outcome, which aims to deliver an AI-model that can better predict recurrence for uLMS patients.