AI-guided sperm retrieval in non-obstructive azoospermia lifts success from 47.6% to 68.1%, with live birth up too
Synopsis
In a retrospective cohort of 1,286 men with non-obstructive azoospermia at two reproductive medicine centers, intraoperative AI-SpermScope marking of likely spermatogenic tubules and rapid sperm confirmation raised sperm retrieval success to 68.1% versus 47.6% for conventional micro-TESE, a gap that persisted after propensity score matching. Among men with retrieved sperm, the subsequent ICSI live birth rate per transfer cycle was 38.5% in the AI group and 26.8% in the conventional group; group assignment was not randomized, so a causal effect still needs prospective randomized trials.
Interpretation
When men with non-obstructive azoospermia undergo micro-TESE, overlaying AI real-time recognition on the surgical microscope yields usable sperm far more often than conventional surgery that relies on the surgeon's eye alone. Conventional micro-TESE sperm retrieval rates are reported at 40% to 60% across centers and depend on surgeon experience and visual discrimination of seminiferous tubules. Among 1,286 patients, retrieval succeeded in 68.1% (369/542) of the AI group versus 47.6% (354/744) of the conventional group, P<0.001; after 1:1 propensity score matching, 486 pairs showed 67.9% versus 48.1%, P<0.001.
The system couples deep-learning image analysis to the surgical microscope, projecting regions of interest likely to contain sperm onto the eyepiece to guide targeted biopsy and quickly confirming sperm in minced tissue. The algorithm was trained on more than 500,000 microscopic images of human testicular tissue and sperm, with previously reported sensitivity of 98.2% and specificity of 96.7% for sperm detection in azoospermic testicular samples. AI-assisted sperm confirmation took about 2 to 3 minutes versus 10 to 15 minutes for conventional manual microscopy; operative duration was 58.32±15.71 minutes versus 87.65±22.43 minutes, and biopsy sites numbered 3.12±1.05 versus 6.87±2.13.
Among men with retrieved sperm who went on to ICSI, the AI group scored higher at fertilization, embryo and pregnancy stages, and multivariable regression tied AI-assisted surgery independently to retrieval success and live birth. Most earlier studies of AI in micro-TESE reported only retrieval rates, with little data on embryo development or live birth after retrieval. The AI and conventional groups showed 2PN normal fertilization of 72.5% versus 64.8%, clinical pregnancy per transfer cycle of 47.2% versus 35.9%, and live birth of 38.5% versus 26.8%, all P<0.05; the multivariable OR was 3.182 (95% CI 2.374–4.265) for retrieval and 1.876 (95% CI 1.325–2.656) for live birth.
Perspective
The results come from 1,286 consecutively enrolled men with non-obstructive azoospermia at two reproductive medicine centers in Sichuan between January 2020 and December 2024; all procedures were performed by the same three surgeons, and group assignment depended on whether the AI-augmented microscope system was available at the time. The authors call for prospective, multicenter randomized controlled trials for further confirmation.
This is a retrospective cohort, and the authors state it cannot establish a causal link between AI assistance and clinical outcomes; patients with sex chromosome aneuploidies, complete AZFa/AZFb deletions or prior testicular surgery were excluded, and 36 men with successful retrieval who went to outside centers for ICSI were left out of the primary live birth analysis; the study did not systematically follow testosterone or testicular volume at 3 and 6 months, nor measure sperm DNA fragmentation.