Infertility affects both men and women, with male infertility often linked to azoospermia, a condition where no sperm is present in semen, making conception impossible. One of the key treatments for azoospermia is testicular sperm extraction (TESE), which involves analyzing histological samples of the testis. These samples are typically scored using the Johnsen score, a scale from 1 to 10 based on the tissue’s appearance.
Aiming to simplify this time-consuming diagnostic process, Dr. Hideyuki Kobayashi, Associate Professor of Urology at Toho University School of Medicine, turned to artificial intelligence (AI). His team used Google’s AutoML Vision, a machine learning platform that requires no programming skills, to create an AI model for analyzing individual patient data. This innovation allows clinicians to build their own AI models without relying on data scientists.
“Our goal was to streamline the diagnosis of azoospermia by leveraging AI,” Dr. Kobayashi said. “With AutoML Vision, clinicians can create and use deep learning models in their daily practice, without needing any specialized programming knowledge.”
The AI model developed by Dr. Kobayashi’s team can analyze histological images of testicular tissue without the need for pathologists. This approach, he believes, can be a game-changer in simplifying clinical practices across various medical fields.
In their study, Dr. Kobayashi and his team defined four categories for Johnsen scores: 1-3, 4-5, 6-7, and 8-10. They used a dataset of 7,155 high-magnification X400 images, which were uploaded to the Google Cloud AutoML Vision platform. The algorithm showed impressive results, with an average precision of 82.6%, a precision rate of 80.31%, and a recall rate of 60.96%.
Despite the growing use of AI in medicine, many clinicians still struggle with applying it effectively due to the need for data science expertise. Dr. Kobayashi believes that this cloud-based machine learning framework can soon become as common in hospitals as software like Microsoft PowerPoint or Excel.
“The most challenging part was collecting the testis pathology images, which was very time-consuming. I want to thank my colleagues for their dedicated efforts in gathering the data for this study,” Dr. Kobayashi added.
This research marks the first successful AI-based algorithm that can predict Johnsen scores accurately without relying on pathologists or data scientists, bringing AI closer to everyday clinical use.
Related topics:
