How machine learning is redefining the “no sperm” diagnosis
For men struggling with infertility, semen analysis is one of the first steps toward finding an answer. About 10 to 15% of these men will be faced with a difficult diagnosis: azoospermia, or the total absence of sperm in the ejaculate. It can feel like a biological dead end, forcing couples to choose between invasive surgery, donor gametes or the end of their biological reproductive journey.
Current guidelines recommend that men with non-obstructive azoospermia undergo microsurgical testicular sperm extraction (mTESE) for sperm retrieval. This is an invasive, and not always successful, surgical procedure where a surgeon searches the testicular tissue for microscopic pockets of sperm production. For the significant proportion of men for whom sperm are not discovered, the result is a significant physical, emotional and financial burden with no reward.
But what if some azoospermic men aren’t actually azoospermic? In a standard semen analysis, only a small portion of the sample is examined for sperm. Certain medical centers have tried to mitigate this limitation with extended sperm searches where embryologists spend up to 8 hours manually searching a single sample, but these analyses are prohibitively expensive and require sustained focus.
In our research, we ask a fundamental question: Are we failing to find sperm simply because they are too rare for the human eye to find? To answer this, we paired high-throughput imaging cytometry, a technology that can image hundreds of cells per second, with a convolutional neural network artificial intelligence model trained to recognize the distinct morphology of human sperm. We analyzed 83 samples that had been clinically labeled “azoospermic” by traditional manual microscopy.
Our platform identified sperm in 48% of those samples, possibly changing the path toward parenthood for these azoospermic patients. Though still in the research and development stage, the potential impact of automated, accurate whole-ejaculate analysis is wide-reaching. Significantly reducing the labor and expertise required, our platform makes exhaustive search scalable and accessible. It gives patients the info they need to make a truly informed decision before moving forward with invasive options such as mTESE, and future work may allow us to isolate these ultra-rare sperm from the ejaculate and avoid surgery altogether.
As this technology moves toward clinical integration, it raises important ethical and policy considerations. The bar for implementation is high, as false positives could lead to costly, unsuccessful IVF cycles, emotional distress and ultimately false hope for patients already going through a stressful period in their lives. Like other advanced reproductive technologies, our platform also prompts questions regarding equitable access. In the current healthcare landscape, IVF and other advanced reproductive technologies often are not covered by insurance, limiting accessibility to those able to pay high out-of-pocket fees. We hope that a more accurate sperm search will lower the overall cost of care, potentially by sparing patients from expensive, unsuccessful surgeries.
By Aidan Boyne, third-year medical student, and Dr. Blair Taylor Stocks, assistant professor in the Division of Male Reproductive Medicine and Surgery in the Scott Department of Urology at Baylor College of Medicine
