Background
Lung cancer is the second most deadly cancer in the Netherlands with more than 10.000 deaths in 2022. Histopathology is at the core of diagnosis and treatment planning but implies time-consuming processes that are becoming increasingly more difficult with the expected shortage on pathologists. Additionally, intraoperatively during the bronchoscopy procedure, tumor detection must be done quickly to optimally guide the next step and avoid longer procedures and increased patient discomfort. To decrease the time needed for tissue sample assessment, higher harmonic generation microscopy (HHGM) might be used. The label-free HHGM images consist of second and third harmonic generation (SHG, THG) and two and three-photon excitation fluorescence (2PEF, 3PEF). SHG shows noncentrosymmetric molecules, such as collagen, THG is generated by local differences of cell and structure interfaces, and 2PEF is produced by endogenous fluorophores such as elastin. 3PEF is hypothesized to show cell cytoplasm and metabolic activity.
Aim
To aid on-site intraoperative tumor detection, we are developing deep learning models for tumor detection and classification, and tissue and cell segmentation. Algorithms will be deployed to the microscope for prospective validation. The algorithms will be trained and evaluated on data resulting from the IMAGIO project.
Funding
This project is supported by the Innovative Health Initiative Joint Undertaking.
