From Resection to Recurrence: Personalizing Glioma Prognosis with a Bayesian Mechanistic and Machine Learning Models.

Datum

07.03.2024

Zeit

14:30 - 16:00

Sprecher

Pejman Shojaee

Zugehörigkeit

Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH) TU Dresden

Serie

IMB - Seminar

Sprache

en

Hauptthema

Medizin

Andere Themen

Biologie, Mathematik

Beschreibung

Glioblastoma (GBM) is an aggressive type of central nervous system cancer with a poor prognosis even with radical treatment. The behavior of cells at their invasive front significantly influences the clinical progression and the quality of life of patients. Additionally, the prediction of tumor behavior or the time frame for recurrence post-surgery is challenging, owing to the tumor's highly heterogeneous nature and the abundance of clinical data. The accuracy in predicting clinical outcomes for this condition is hindered by two main issues: first, a limited understanding of the fundamental mechanisms controlling the data variables, and second, inadequate data collection resulting from dependence on the patient's clinical symptoms. The first issue affects the precision of mechanistic models and precise underlying mechanisms, and the second issue limits the ability of machine learning algorithms to accurately deduce the dynamics of the disease. To address these challenges and achieve a more accurate result as the time to relapse, we utilize patient-specific MRI data as input to derive key modelable parameters, including tumor growth and infiltration rate, tailored for each patient and biopsy data to extract the number of tumor-associated macrophages (TAMs) and tumor cells. These parameters are then integrated into our mathematical models as an informative Bayesian prior. Furthermore, we incorporated clinical and radiomics data to enhance the non-modelable component of our approach, which involves the application of machine learning methods. In the end, we develop a Bayesian combination of mechanistic modeling and machine learning algorithms to increase the accuracy of our patient-specific predictions. This method enables us to determine the probability distribution of recurrence timing for each patient, which can be advantageous in tailoring the second line of treatment individually.

Links

Letztmalig verändert: 19.02.2024, 10:52:01