Automatic directional masking technique for better sperm morphology segmentation and classification analysis

Abstract

Male-based factors have very critical effects in infertility. The determination of male factors is based on the semen analysis performed by multiple aspects such as motility, morphological and concentration. Normally, tests are applied by experts manually. Therefore, the results are highly subjective. Automation of analysis minimises the human factor. In this respect, a number of studies on the computerised semen analysis have been recently reported. One of the most important steps in computerised approaches is the region of interest extraction. In this Letter, a novel masking technique is introduced with the aim of better region of interest extraction, in which each sperm is individually analysed by head and tail specified stages. In this approach, wavelet-based image enhancement techniques and gradient analysis are jointly utilised for finding the sperm orientation. Later, elliptic masks are employed for the sperm segmentation along this orientation. As the evaluation methods, firstly proposed technique's segmentation performance is assessed by visual inspection. Secondly, the k-nearest neighbours (k-NN) classification results of proposed approach's outputs are compared with the outputs of classical k-means segmentation and also the raw images. Results indicate that 93.5% of all the sperms are correctly segmented by proposed masking approach and up to 13% increase is observed in k-NN classification.