Master / Bachelor Thesis : Foundation Model Empowered Computational Pathology Images Analysis

Master / Bachelor Thesis : Foundation Model Empowered Computational Pathology Images Analysis

27.05., Studentische Hilfskräfte, Praktikantenstellen, Studienarbeiten

A Master / Bachelor thesis opportunity to evaluate foundation model performances on downstreaming pathology image tasks.

Artificial intelligence (AI) can potentially transform cancer diagnosis and treatment by analyzing pathology images for precision medicine and decision support systems. Pathology’s clinical practice usually encompasses tasks like tumor classification, segmentation, subtyping, grading, staging, and whole slide matching. Although AI demonstrates promise in many pathological tasks, it still faces challenges in generalization and addressing rare diseases due to limited training data availability. [CDL+24, VBC+23]

Here, a foundation model may contribute to this challenge. A foundation Model refers to a general- propose model pre-trained on typically unlabeled datasets, subsequently fine-tuned to apply to diverse downstream tasks [DFW+24]. To compare the proposed foundation model with previous state-of-the- art methods, we want to evaluate the performance of patch / slide level classification and segmentation tasks.

Tasks :

• Paper reading and literature review
• Evaluate proprietary and public foundation models on public datasets
• Finetune the models to obtain the best performance
• Publicly available models : UNI [CDL+24], cTransPath [WYZ+22], Virchow [VBC+23], RudolfV [DFW+24], MSSM [CKF+23]
• Datasets for classification and detection : MHIST [WSR+ 21], PCAM [VLW+ 18], NCT-CRC [KHM18], CAMELYON16 / 17 [BGM+18]
• Datasets for segmentation : SegPath [KOS+23], PanNuke [GAKB+19]

Requirements :

Basic knowledge in at least one of the following areas :

Supervision and Contact :

Prof. Peter Schueffler and Jingsong Liu will be the supervisors.