Principal Investigator Charles Sodini
Urinalysis is one of the most common diagnostic tech- niques in medicine. Over 200 million urine tests are ordered each year in the US, costing between $800 to $2,000 million in direct costs. 46% of all urinalysis tests include microscopic analysis, which involves identifying and counting each particle found in the urine. Microscopic urinalysis is a costly and complex process often done in medical laboratories. An inexpensive and automated cell-counting system would (1) increase access to microscopic urinalysis and (2) shorten the turn-around time for physicians to make diagnostic decisions by permitting the test to be done at the point- of-care.
The AutoScope is an automated, low-cost microscopic urinalysis system that can accurately detect red blood cells (RBCs), white blood cells (WBCs), and other particles in urine. We use a low-cost image acquisition system combined with two neural networks to identify these particles. By not using any optical magnification, we achieve costs three orders of magnitude less than the only commercially available semi-automated urinalysis system and a device size of 8.3 x 6.0 x 8.8cm.
To validate the system, we calculated the accuracy, sensitivity, and specificity of the Autoscope. The specificity and sensitivity were determined by generating 209 digital urine specimens modeled after urine received in medical labs. The Autoscope had a sensitivity of 88% and 91% and a specificity of 89% and 97% for RBCs and WBCs, respectively. Next, we determined the Autoscope’s accuracy by fabricating 8 synthetic urine samples with RBCs, WBCs, and microbeads. The reference results were confirmed through a medical laboratory. The AutoScope’s counts and the reference counts were linearly correlated to each other (r2= 0.980) across all particles. The sensitivity, specificity, and R-squared values for the AutoScope are comparable (and mostly better) than the same metrics for the iQ-200, a $100,000-$150,000 state-of-the-art semi-automated urinalysis system.