Sperm Hunter - Neural Network Model for Sperm Identification in Azoospermic Males

Azoospermia is defined as a lack of sperm in the ejaculate and causes infertility in 1% of men worldwide. For 60% of these men, the lack of sperm is caused by spermatogenic dysfunction (ASD) where the testis is unable to produce sperm as opposed to a blockage in the ejaculatory pathway. Microdissection testicular sperm extraction (MicroTESE) is the recommended treatment to extract sperm directly from the testis for use in IVF, however, the surgery is not as accurate as it can be, and every failure costs a potential human life. A novice surgeon, prior to their first 50 procedures, will have a 12% lower sperm retrieval rate and will perform the procedure 25 minutes slower than an expert surgeon. The learning curve for the procedure continues to be steep as even an expert surgeon will reach 50% average sperm retrieval only after 250 procedures. To address this problem, we created a rodent model of ASD using the spermatotoxic chemotherapy agent busulfan. A neural network was trained on images taken during the MicroTESE procedure on ASD rats and sperm counts associated with the sampling locations to develop a model to detect sperm-dense locations. The resulting model achieved sperm prediction rates comparable to fellowship-trained surgeons when evaluated by an operator with moderate medical training and no prior experience with the MicroTESE procedure. Once validated in human studies, this model has the potential to augment sperm retrieval rates and deliver reproducible results during MicroTESE procedures.