Sparse Optical Acquisition for Spatial Spectral Reconstruction in Hyperspectral Microscopy
FOTONİK 2026 | Ulusal Optik, Elektro-Optik ve Fotonik Çalıştayı, Ankara, Türkiye, 11 Eylül 2026, cilt.1, ss.19, (Özet Bildiri)
- Yayın Türü: Bildiri / Özet Bildiri
- Cilt numarası: 1
- Basıldığı Şehir: Ankara
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.19
- Kocaeli Üniversitesi Adresli: Evet
Özet
Hyperspectral microscopy integrates spatially resolved optical imaging with wavelength resolved spectral measurements, enabling characterization of wavelength dependent light matter interactions in biological samples. Variations in absorption, scattering, and reflectance across the spectrum generate characteristic spectral signatures that can improve discrimination of cellular and tissue structures beyond conventional intensity based microscopy. However, sequential raster based acquisition produces large hyperspectral data volumes and relatively long measurement times, which may restrict the spatial area that can be practically examined.
In this study, systematic and stratified random acquisition strategies were investigated to reduce the number of physically acquired hyperspectral measurements while preserving the wavelength dependent optical response, spatial information, and spectral fidelity required for reconstruction. A fixed prostate cancer cell line sample in a single Petri dish was first imaged using complete raster scanning to obtain a reference hyperspectral dataset. Without changing the sample position, partial datasets were subsequently acquired at five sampling ratios ranging from 20% to 60%, using systematic and stratified random sampling with three sampling seeds (42, 111, and 2026). Missing spatial spectral information was reconstructed using nearest neighbor and linear interpolation. The reconstructed hypercubes were compared with the full raster reference using PCA based spatial analysis and spectral similarity metrics, including spectral angle mapper (SAM), R², RMSE, spectral correlation, and Euclidean distance.
Across the evaluated sampling conditions, reconstructed datasets showed high agreement with the raster reference, with R² values of 0.91–0.95, spectral RMSE of 0.097–0.133, spectral correlation of 0.955–0.978, and mean SAM values of 4.25–6.24°. These results indicate that wavelength dependent spectral characteristics and spatial spectral structure can be successfully preserved after reconstruction over sampling ratios of 20–60%. The proposed framework therefore offers a promising approach for reducing optical acquisition burden while maintaining spectral integrity in biological hyperspectral microscopy.