A novel endoscopic spectral imaging platform integrating k-means clustering for early and non-invasive diagnosis of endometrial pathology.

We report a multimodal endoscopic system capable of performing both color and fast multispectral imaging in the spectral range 400-1000 nm. The system is based on a computer controllable tunable light source, which can be coupled with all types of endoscopes. Performance evaluation showed about 60% flat transmittance in almost all the operating wavelengths, at about 13 nm bandwidth per tuning step. With this system adapted to a thin hysteroscope, we also report, for the first time, spectral analysis of the endometrium and unsupervised/objective clustering of the spectra. We have implemented a method combining the k-means algorithm with the silhouette criterion for estimating the number of the distinguishable spectral classes that may correspond to different medical conditions of the tissue. It was found that there are five-well defined clusters of spectra, while preliminary clinical data seem to correlate well with the tissue pathology.