A Learning-Based Framework for the Automatic Segmentation of Human Sperm Head, Acrosome and Nucleus

Abstract

Evaluating the morphology of the human sperm is one of the most important steps in the human semen analysis, which is the controversial aspect of the treatment of male infertility. Manual assessment of the corresponding parameters of human sperm morphology is a time-consuming, reader subjective and error-prone process. Therefore, developing the automatic methods is necessary to achieve the more accurate diagnosis. In this paper, we presented a learning-based framework for the automatic segmentation of the human sperm head particles, i.e., Acrosome and Nucleus. First, the homomorphic filtering is employed to correct uneven illumination and highlight each sperm from the background of an image. In the second step, sperm's heads are segmented using an introduced deep convolutional neural network (CNN). Then, a filling holes operation and geometric constraints are utilized to improve head segments. A Support Vector Machine (SVM) is used to classify each pixel of segmented heads to nucleus and acrosome regions. Finally, segmented acrosomes and nucleus are modified using opening and closing operations followed by isolated objects removing. The proposed method is validated on the expert delineated dataset with 20 images of human semen smears and obtains 0.94, 0.87, and 0.88 of Dice Similarity Coefficient for the head, the acrosome, and the nucleus segments, respectively. Our results indicate that the proposed method has outperformed the segmentation systems based on classical learning methods, in the accuracy and the reliability.