AI sperm-selection research explores whether microscopic images can help estimate DNA integrity beyond conventional visual assessment. Early studies show laboratory prediction or assay-analysis capabilities, but a predicted score is not proof that a sperm is genetically normal. The separate question is whether using the tool improves pregnancy and live-birth outcomes.
In intracytoplasmic sperm injection (ICSI), an embryologist selects an individual sperm for injection into an egg. Morphology and motility inform this choice, but visual selection does not directly measure every aspect of DNA quality. This is the gap explored by image-based prediction research.
Research asks whether computational analysis can add information that is difficult to judge by eye. That aim should be distinguished from a claim that AI selection has already replaced conventional practice.
What AI sperm-selection systems aim to estimate
For example, McCallum and colleagues’ 2019 study trained a neural network using 1,064 sperm images from 6 healthy donors. It predicted a single-cell DNA-integrity measure from bright-field images with a reported correlation of about 0.43. This is probabilistic prediction, not confirmation that a sperm carries no genetic abnormality.
Morphology and DNA integrity are different questions
Morphology, sperm DNA fragmentation and chromosome status are different questions. The EAU guideline describes associations between DNA fragmentation and reproductive outcomes, but also uncertainty in predicting assisted-reproduction results. High DFI does not establish the cause of an individual fertilization failure, embryo arrest or failed implantation.
AI also appears in laboratory assay analysis. A 2023study by Kuroda and colleagues compared AI-assisted sperm chromatin dispersion scoring with manual and other assays in 20 men. Agreement with manual scoring was strong (r = 0.97), but differences between assay types remained. That is an assay-comparison result, not a trial selecting living sperm to improve birth outcomes.
These tools could help investigate a sperm-related part of reproduction, but they do not establish that sperm defects were the overlooked cause of every unsuccessful IVF cycle. Embryo development, uterine factors and both partners’ circumstances still require consideration.
For unexplained infertility, recurrent implantation failure or recurrent pregnancy loss, the clinical question is what an additional test or selection method would change. AUA/ASRM guidance supports selected uses of DNA-fragmentation testing, including recurrent pregnancy loss; that recommendation does not itself validate an AI selection product.
Validation gaps: standards and live-birth outcomes
Validation remains essential. A useful tool must perform across patients, laboratories and imaging conditions, and a clinical study must test whether using it improves meaningful outcomes. The studies described here do not establish a consistent live-birth advantage. Their laboratory results should not be presented as such a benefit.
AI may eventually support an embryologist’s judgment by adding validated information. For now, predicting DNA integrity, measuring fragmentation in an assay and showing a treatment benefit are separate achievements. Progress in the first two is a reason for further evaluation, not proof that the cause of IVF failure has been solved.
Research sources and limits
McCallum et al., Communications Biology (2019), doi:10.1038/s42003-019-0491-6, University of Toronto and collaborating fertility researchers: small laboratory prediction dataset, not a pregnancy trial. Kuroda et al., Andrology (2023), doi:10.1111/andr.13436: 10 donors with normal semen and 10 infertile men, comparing DNA-fragmentation assays. These examples concern different AI applications and should not be treated as interchangeable clinical tools.
Related reading
- Sperm DFI and IVF versus ICSI: a study of cumulative live birth
- ASFI versus ICSI: a different approach to assisted fertilization
Disclaimer: This content is provided for informational purposes, based on reporting on infertility and various public data. Medical judgments and treatment decisions must always be made in consultation with professional medical personnel. Image: AI-generated (ChatGPT, OpenAI) / Visual reference for illustrative purposes only.