Fast and Accurate Sperm Detection Algorithm for Micro-TESE in NOA Patients

Non-obstructive azoospermia (NOA) presents major challenges in assisted reproductive technology (ART) due to the extremely low number of viable sperm within testicular tissue. In Micro-TESE procedures, embryologists manually search for sperm under DIC microscopy-a slow, labor-intensive process. We aim to streamline this process with an efficient computational detection tool.

We present SD-CLIP (Sperm Detection using Classical Image Processing), a lightweight, real-time algorithm that simulates sperm structure detection from unstained DIC images. The model first identifies convex sperm head candidates based on shape and width using edge gradients, then confirms the presence of a tail via principal component analysis (PCA) of pixel clusters.

Compared to the MB-LBP + AKAZE method, SD-CLIP improved processing speed by 4× and achieved a 3.8× higher posterior probability ratio, making detected sperm candidates significantly more reliable. Evaluation was performed on both human Micro-TESE and mouse testis images, demonstrating robustness in low-sperm environments.

SD-CLIP simulates a domain-specific image interpretation model that identifies sperm morphology with high specificity. It requires minimal computational resources, supports real-time integration, and could be extended to automated sperm extraction systems. This tool has clinical value for accelerating Micro-TESE and increasing success rates in ART for NOA patients.