The team led by Dr. Lin-shien Fu, head of the Pediatric Medicine Center at Taichung Veterans General Hospital, has developed multiple artificial intelligence modules for automatically identifying abnormalities in kidney ultrasound images based on the hospital's extensive database of kidney ultrasound images. These modules are combined to establish an auxiliary diagnosis system for pediatric kidney ultrasound image abnormalities.

Accurately identifying lesions such as hydronephrosis and cysts in images

Kidney diseases often lack obvious symptoms due to their chronic nature, making them easy to overlook. Common physical examinations, including urine and blood tests, still miss many kidney problems. Traditional manual interpretation of kidney ultrasound images is often limited by the subjective judgment and experience of physicians, making it difficult for non-nephrologists to determine the interpretation results. To improve the accuracy of kidney ultrasound image interpretation, the team selected tens of thousands of original ultrasound images from a large database and established multiple abnormal image classification training sets, including hydronephrosis, cysts, stones, hyperechogenicity, and space-occupying lesions. Using transfer learning based on the ImageNet dataset, the team repeatedly tested the optimal fitting state of the last four layers' recognition ability and quantitatively verified the model's performance through model evaluation, providing visualizations for expert interpretation.

The diagnostic system has the following advantages:

• High sensitivity and accuracy: Sensitivity of 96.36% and accuracy of 92.5%, effectively screening kidney ultrasound image abnormalities.
• Rapid interpretation: Processing 100 ultrasound images to generate reports takes less than 30 seconds, and transmission of kidney ultrasound images between Taiwan and Vietnam for interpretation can be completed in less than two seconds, providing physicians with real-time auxiliary diagnostic results.

1. Applicable to rural and international settings: Interpretation can be performed via the internet, suitable for areas with limited medical resources, even for international use.

Harnessing artificial intelligence to achieve healthcare accessibility

Dr. Lin-shien Fu pointed out that this technology has not only been successfully published in international journals but has also obtained patents from the Republic of China. It is currently undergoing the process of applying for a U.S. patent, with an early publication in February 2024. By popularizing this AI-assisted diagnostic system in primary care clinics, health check-ups, and even in remote areas and internationally, it is expected to significantly increase the detection rate of kidney diseases in children and even across all age groups, providing people with more convenient medical services.

Resource (mandarin):

AI揪兒童腎臟異常！超音波影像判讀助力基層偏鄉篩檢