Artificial intelligence will accelerate the evaluation process of testicular biopsy results

The problem of azoospermia and its diagnosis

Azoospermia, which refers to the total absence of sperm in the ejaculate, is a serious issue that prevents the conception of a child. In the treatment of couples where the male has this diagnosis, testicular biopsy can be used to obtain sperm. The samples obtained are evaluated using the Johnsen scale, which provides a grading from 1 to 10 points.

Introduction of AI in histological evaluation

“For over 50 years, the Johnsen scale has been used in urology. However, the histological evaluation of testicular tissue is not an easy task and requires a lot of time. We sought to simplify this time-consuming diagnostic step by implementing artificial intelligence (AI) technology. The Google Automated Machine Learning (AutoML) Vision software was chosen to create the AI model. The AutoML Vision technology can be used by healthcare professionals without programming skills,” said the head of the research team, Hideyuki Kobayashi, associate professor of urology at Toho University (Japan).

Grading of the scale and model training

To simplify the application of the Johnsen scale, Dr. Kobayashi divided it into four groups: 1-3, 4-5, 6-7, and 8-10 points. This grading reduces the number of categories and facilitates the automatic classification of images while preserving the clinical significance of the differences between levels. As part of the study, 7155 images of histological specimens were uploaded to Google’s Cloud AutoML Vision platform. The artificial intelligence demonstrated an accuracy of 80% in evaluating the histological model of the specimens.

Prospects for the use of AI in medicine

AI is gaining popularity in various fields of medicine. For example, AI-based technologies have been successfully used to analyze images for breast cancer diagnosis and to detect pathologies in pulmonary CT images. However, its large-scale use in clinical practice is still hindered by the need for data processing specialists.

“The machine learning technology we chose is aimed at medical personnel. It is likely that in the near future, thanks to AI, doctors will use medical image classifiers with the same ease and accessibility as they currently use standard office software tools,” Dr. Kobayashi stated.