Development of mosaic camera based hyperspectral dermatoscope

Cancer margin detection is of vital importance for successful treatment. This study aims to develop image demosaicing methods for mosaic camera based hyperspectral dermatoscope (HSd). Such a camera is built by attaching a periodic grid pattern of filters to an imaging sensor, resulting in each pixel detecting a specific narrow wavelength range (FWHM 15 nm). In previous works a single grid was converted to a single pixel in the hypercube. This leads to a significant reduction in spatial resolution, which we intend to resolve.

HSd is based on the HS03-G2 mosaic camera (Photonfocus AG, Lachen, CH). For the optics we use a telecentric lens with a field of view of around 2 by 2 cm. Samples were illuminated with custom LED ring light.

Because previous work has shown that filter response can vary strongly between multiple pixels with the same theoretical wavelength range, we performed spectral characterization of each pixel using bandpass filters. Afterward, demosaicing algorithms were developed to fill in the missing spectral data. Validation was first done using color checkers and then human subjects by performing the cuff test. Captured images were compared using a spectral angle mapper method to a well characterized HSI system.

Preliminary results show good correspondence between images captured using HSd and characterized HSI system, while still retaining the full resolution of the imaging sensor. High resolution real-time HSI could provide better diagnosis and unlock the potential for image-guided surgery.