The Open Kidney Ultrasound Data Set

Rohit Singla, Cailin Ringstrom, Grace Hu, Victoria A. Lessoway, Janice Reid, Christopher Y. Nguan, Robert Rohling

Abstract: Ultrasound is widely used and affordable diagnostic tool for medical imaging. With the increasing popularity of machine learning, ultrasound research has also expanded, but limited access to open data sets hinders progress. Despite being a frequently examined organ, the kidney lacks a publicly available ultrasonography data set. To address this issue, the proposed Open Kidney Ultrasound Data Set is the first publicly available, high-quality data set of kidney B-mode ultrasound data, collected over five years from over 500 adult patients with common primary diseases. The data set includes annotations for multi-class semantic segmentation, with fine-grained manual annotations from two expert sonographers. It contains images from a patient population with a mean (± stdev) age of 53.2 ± 14.7 years, with 63% males and 37% females. The primary diagnoses included include diabetes mellitus (32%), immunoglobulin A nephropathy (11%), hypertension (10%), and other diseases (\(\le 10\%\) each). Of the images, 91.2% were rated as fair or good quality, while 8.8% were deemed poor or unsatisfactory. Intra-rater and inter-rater variability were assessed, and initial benchmarking demonstrated a state-of-the-art algorithm achieving a Dice Sorenson Coefficient of 0.85 for kidney capsule segmentation. Furthermore, the data set includes both native and transplanted kidneys, providing a comprehensive data set for future researchers to develop novel image analysis techniques for tissue characterization, disease detection, and prognostication.