Database of Pediatric Kidney Scans for AI Projects
A University of Pittsburgh researcher has developed a database of renal scans from children with detailed diagnostic information. These scans, each assessed by three independent radiologists and their associated detailed reading of each kidney segment, clinical data, and demographic information, could form a valuable training set for artificial intelligence (AI) algorithms to automate assessment of kidney scans. Development of AI-based technology to review and assess kidney scans could dramatically improve diagnosis efficiency, lead to prognosis predictions that allow clinicians to better tailor therapy, improving outcomes in patients.
Description
AI is increasingly used to support clinical decision making, assisting medical professionals to make the best decisions for their patients. Of particular interest is using AI to assess clinical images and is currently used in diagnostic workflows in oncology, hematology, and dermatology. The development of AI algorithms to assess images requires large training sets of images with associated detailed interpretation from clinicians and patient data. This database, consisting of approximately 300 renal scans from children is unusually large. With independent analysis from three radiologists, it could provide a rich source of training data for the development of an AI algorithm to assess renal scans in children.Applications
• Development of diagnostic AI-based software• Development of clinical decision-making AI-based software
Advantages
Diagnosis of kidney disease in children can be complex. Symptoms can be subtle, and some children display no symptoms at all. Kidney disease can impact on growth and development, including language and behavioral development, and creating tools for more efficient diagnosis of kidney disease should lead to earlier diagnosis and treatment. Currently, invasive biopsies are often required for diagnosis. Development of accurate diagnosis based on non-invasive imaging would remove the need for biopsies, reducing biopsy-associated risks including infection and pain.This novel database of images from pediatric patients contains demographic and clinical details of patients along with detailed radiologist readings and is unusually large. The rich dataset, which includes information about each segment of the kidney, could be used to train AI algorithms to assess kidney scans and produce detailed observations rapidly. These observations could assist in diagnosis and clinical decision making for pediatric nephrologists.