Nuclei location enhancement based on improved efficient ellipse hough transform for third harmonic generation microscopy imaging

Instead of conventional manual cell segmentation, this paper introduces a computer-aided diagnosis protocol based on non-invasive in vivo virtual biopsy, third harmonic generation (THG) microscopy, which provides consistency and prompt suggestions for immediate clinical decisions without pain to patients. Specifically, we present a nuclei location re-inspection mechanism using enhanced efficient ellipse Hough transform with labeling and weighting information building upon previous watershed-based nuclei segmentation. This transform approach cautiously identifies nuclei location by considering not only staining intensity, but also examine shape feature to enhance the accuracy of nuclei recognition. The current approach has been implemented and tested on about 400 THG images from 25 healthy volunteers, and the receiver operating characteristic (ROC) curve has demonstrated the better performance of decreased false positive rate and increased true positive rate as compared to other nuclei location detection algorithms.