Predicting the primary tumor of brain metastases using simulated Raman Histology

Current brain tumor diagnosis is time-consuming and depends on multiple clinical experts, making intraoperative diagnosis crucial for safe and efficient brain surgery. Raman histology, a new stain-free optical imaging method, significantly reduces tissue imaging time while capturing key morphological elements similar to hematoxylin and eosin staining.

We explore the potential of Raman imaging to extract critical diagnostic information about the cancer origin using machine learning methods.

We will develop and implement image analysis pipelines, designing methods to predict tumor type from images using techniques such as convolutional neural networks, weakly supervised methods, and transformers. Additionally, interpretability mechanisms will be created to highlight informative image regions for predictions. Medical experts will collaborate to interpret the visual features utilized by machine learning models.

This project aims to improve intraoperative brain tumor diagnosis. Accurate identification of primary tumor type is essential for optimal surgical strategies. Raman imaging combined with automated image analysis could drastically reduce diagnostic time, enhancing patient safety.