CS3: Cascade SAM for Sperm Segmentation

TL;DR

CS3 tackles the critical problem of segmenting individual sperm in overlapping images for automated morphology analysis. It introduces Cascade SAM (CS3), an unsupervised pipeline that applies multiple SAM passes in a cascade to segment heads, simple tails, and complex tails, followed by mask matching to assemble complete sperm masks. The method leverages color-based head filtering, skeletonization-based tail filtration, and end-point geometry with thresholds ($\lambda_{dis}=20$ px, $\lambda_{angle}=60^\circ$) to pair head and tail masks without labeled data. Experimental results on clinical data (≈2,000 unlabeled images and 240 expert-annotated images) show CS3 achieves higher mean IoU ($mIOU$) and mean Dice ($mDice$) than supervised and unsupervised baselines, particularly for overlapping sperms. The work enables scalable automated sperm morphology analysis and suggests applicability to segmentation of other elongated biological structures.

Abstract

Automated sperm morphology analysis plays a crucial role in the assessment of male fertility, yet its efficacy is often compromised by the challenges in accurately segmenting sperm images. Existing segmentation techniques, including the Segment Anything Model(SAM), are notably inadequate in addressing the complex issue of sperm overlap-a frequent occurrence in clinical samples. Our exploratory studies reveal that modifying image characteristics by removing sperm heads and easily segmentable areas, alongside enhancing the visibility of overlapping regions, markedly enhances SAM's efficiency in segmenting intricate sperm structures. Motivated by these findings, we present the Cascade SAM for Sperm Segmentation (CS3), an unsupervised approach specifically designed to tackle the issue of sperm overlap. This method employs a cascade application of SAM to segment sperm heads, simple tails, and complex tails in stages. Subsequently, these segmented masks are meticulously matched and joined to construct complete sperm masks. In collaboration with leading medical institutions, we have compiled a dataset comprising approximately 2,000 unlabeled sperm images to fine-tune our method, and secured expert annotations for an additional 240 images to facilitate comprehensive model assessment. Experimental results demonstrate superior performance of CS3 compared to existing methods.

Figures (4)

• Figure 1: Preliminary Investigations into SAM on pre-processed sperm images. (a)-(b) After removing sperm heads, SAM begins to segment remaining parts from a geometric perspective. (c)-(d) After removing simple parts, SAM begins to focus on complex areas. (e)-(f) After enlarging and bolding, overlapping parts that are difficult to split become separatable.
• Figure 2: The architecture of our method CS3. Through the cascade application of SAM, sperm heads, simple tails, and complex tails are segmented step by step. All masks are matched and spliced to generate complete masks. $S_{1}, S_{2}, \cdots, S_{n}$ donotes a series of SAM in everything mode.
• Figure 3: Segmentation results of CS3. (a) The raw images. (b)-(e) The results of cascade SAM gradually segmenting different parts of sperm from simple to complex. (f) The results after mask matching and splicing. More results can be found in the appendix.
• Figure 4: Segmentation results of different methods. More results can be found in the appendix.