AI looking handy with 3D abdominal ultrasound

"Mayo Clinic researchers have demonstrated a deep learning model that can automatically segment kidneys and measure total kidney volumes using only 3D ultrasound images.

While the team concentrated on using the technique to assess a relatively uncommon condition—an inherited cyst-forming disorder called autosomal-dominant polycystic kidney disease (ADPKD)—they suggest the AI approach may have broader clinical implications.

Co-corresponding authors Bradley Erickson, MD, PhD, Timothy Kline, PhD, and colleagues trained the AI, a convolutional neural network, on multiple kidney ultrasound scans from 22 patients who had ADPKD. They tasked the tool with segmenting whole kidneys and measuring total kidney volumes.

For comparison purposes they also imaged the patients with the gold-standard modality, abdominal MRI, and pitted the AI against radiologists with expertise in abdominal image annotation and segmentation.

Reporting the results April 27 in Abdominal Radiology, the authors describe the neural network’s performance as “promising for auto-segmentation of kidneys and calculating total kidney volume,” adding that its yield was close in quality to the highly labor-intensive manual tracings and measurements traditionally performed by radiologists.

Further, with the help of AI, the ultrasound images compared well with those from the MRI scans, “suggesting [the approach] may be useful in populations where MRI is more challenging, such as children.”..." Lire la suite