Deep‐Learning‐Based Profiling of Mouse Sperm Head and Neck Morphology Reveals Coordinated Structural Variation
Gwidong Han, Seung Pyo Hong, Seung Jae Lee, Chunghee ChoABSTRACT
Background
Mammalian sperm head morphology varies across species and dictates fertilization capability. Rodent sperm have asymmetric, hook‐shaped heads, complicating analysis. Despite widespread biomedical use, current tools poorly characterize these complex variations.
Objectives
This study aimed to establish a robust framework for the precise, noninvasive, and quantitative analysis of mouse sperm morphology.
Materials and Methods
We developed ADAM‐net (Anomaly‐aware Deep‐learning Architecture for Morphology), a machine‐learning system for mouse sperm head and neck structures. Integrating a dataset preparation module, ResNet‐18, and additional layers, it features two variants: ADAM‐net‐FL for fluorescence (FL) and ADAM‐net‐BF for bright‐field (BF) images.
Results
ADAM‐net‐BF enables simultaneous analysis of the head and neck using noninvasively acquired BF images. Atypical head shapes were frequently associated with a neck in which the tail is attached perpendicular to the head axis, revealing a specific relationship between head shape and tail attachment pattern. Additionally, ADAM‐net provided novel quantitative insights into testicular germ cell‐specific HSF2‐interacting lncRNA ( Teshl )‐knockout sperm defects, supporting its utility for mouse sperm morphology analysis.
Conclusion
ADAM‐net provides a standard and facile tool that will have applications in various reproductive studies, extending to the exploration of subtle structural dynamics such as the sperm head–tail alignment.