Speaker
Description
This work focuses on hyperspectral imaging in biomedical applications, particularly fluorescence microscopic hyperspectral imaging (FMHSI) of unstained biological tissue, a foundational tool in diagnostic pathology and biomedical research due to its high spectral and spatial resolution. Semantic segmentation of FMHSI images is essential for generating labeled HSI data and enabling downstream quantitative analysis.
We develop an automatic unsupervised semantic segmentation framework that produces high-quality segments for high-level vision tasks such as disease diagnosis and pathology-assisted surgery. The proposed multi-step algorithm includes denoising and dimensionality reduction, hierarchical clustering-based segmentation using spectral and spatial information, and adjacent region merging for refinement. The algorithms will be open-sourced and integrated into an application to facilitate analysis of FMHSI data, with potential applications in improving the understanding and diagnosis of eye diseases.
