ON THIS PAGE:
Related clinical information & resources
Related health professional education
Machine learning-based prediction models in male reproductive health: Development of a proof‐of‐concept model for Klinefelter Syndrome in azoospermic patients. Krenz et al., Andrology 2022
When it comes to diagnosing melanoma, we’re now at a point where the most experienced dermatologists can be outperformed by computers. As large numbers of patient records become digitised, it’s become possible to use the swathes of clinical data and diagnoses to train computers to identify patterns associated with disease.
Machine learning has been applied to semen analysis to try and automate some of the process, and perhaps predict which patients might be good candidates for specific assisted reproductive techniques or be likely to need further investigations.
Infertility due to azoospermia may be caused by several different factors. For about one-quarter of azoospermic men, the cause is some form of obstruction in the seminal tract. For the remaining 75% of men with non-obstructive azoospermia, a common cause is Klinefelter syndrome.
Somewhere between 7.5%-10% of men with non-obstructive azoospermia have Klinefelter syndrome. Testicular dysgenesis and atrophy, which cause the hallmark small testicular volumes of the 47,XXY karyotype, prevent natural conception for men with Klinefelter syndrome.
It’s possible to recover viable sperm using testicular sperm extraction, with a success rate of around 45%. The use of these sperm for intra-cytoplasmic sperm injection (ICSI) can result in viable embryos and successful pregnancies for nearly half of the people from whom sperm are collected.
Machine learning has been applied to successfully predict the likelihood of Klinefelter syndrome in men with azoospermia, using routine clinical data. The result is an algorithm that correctly classifies all azoospermic men with Klinefelter syndrome, based on six factors: age, height, total testis volume, ejaculate volume, and serum concentrations of LH and FSH. Total testis volume was the most important of these factors in the prediction algorithms.
Algorithms generated by machine learning correctly identified all individuals with Klinefelter syndrome in a large set of data from azoospermic men, outperforming specialist urologists and andrologists4. Expert humans were better, however, at classifying men without Klinefelter syndrome correctly.
Karyotyping is the gold standard for diagnosis of Klinefelter syndrome, but the application of an algorithm for diagnosing Klinefelter syndrome in men with azoospermia could result in faster diagnosis, better clinical care and, ultimately, parenthood for otherwise infertile couples.
The best performing algorithms from the research described above have been applied to a ‘Klinefelter Score Calculator’ available (for research purposes) here.
 Haenssle et al., 2018. Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Annals of Oncology
 Kasak & Laan, 2021. Monogenic causes of non-obstructive azoospermia: challenges, established knowledge, limitations and perspectives. Human Genetics
 Corona et al., 2017. Sperm recovery and ICSI outcomes in Klinefelter syndrome: a systematic review and meta-analysis. Human Reproduction Update
 Krenz et al., 2022. Machine learning based prediction models in male reproductive health: Development of a proof‐of‐concept model for Klinefelter Syndrome in azoospermic patients. Andrology