Deep Bio’s prostate cancer AI algorithm shows high accuracy in prostate cancer diagnosis: study

Deep Bio, a Korean AI-based digital pathology and diagnostic company, said Tuesday it has published a study demonstrating the efficacy of its prostate cancer AI analysis algorithm, DeepDx Prostate, in collaboration with Stanford Medicine. 

The study, "External validation of an artificial intelligence model for Gleason grading of prostate cancer on prostatectomy specimens," was published on July 12 in the British Journal of Urology International established in 1929 by the British Association of Urological Surgeons.

The study evaluated the performance of the DeepDx Prostate AI algorithm in detecting and grading prostate cancer in prostatectomy specimens. Trained on core needle biopsy (CNB) images collected from two hospitals in Korea, the algorithm was tested on specimens from different institutions and demonstrated “outstanding results” without additional fine-tuning, Deep Bio said.  

According to the study, the algorithm achieved a sensitivity of 0.997 and a specificity of 0.88 in detecting cancer in prostatectomy specimens. It also showed high concordance with urological pathologists, with a Cohen's kappa of 0.91 for cancer detection, 0.89 for Gleason grading system, and 0.89 for identifying risk groups (normal or low-risk [GG1], intermediate-risk [GG 2-3], and high-risk [GG 4-5]).

Deep Bio CEO Kim Sun-woo said the pathological examination of prostatectomy specimens required “significant time and effort.” 

Despite these challenges, the DeepDx Prostate AI algorithm, although trained on biopsy specimens from a different patient set, performed well in both identifying and grading cancer on prostatectomy (RP) specimens. The algorithm’s results closely matched those of expert uropathologists, demonstrating its robustness and reliability across varied patient groups and specimen types, the company said.

Dr. Bogdana Schmidt, the study's lead author, a former Stanford researcher and now an associate professor and urologic surgeon at the University of Utah, said, “DeepDx Prostate’s key functionality in rapidly and accurately grading prostate cancer demonstrated the potential for its AI algorithm to be widely adopted in clinical practice, assisting in the development of effective treatment plans.”

A Deep Bio official said that a cancer diagnostics company, which she did not reveal at the company's request due to sensitivity, will oversee the sales of the DeepDx Prostate AI algorithm in Korea. 

"A patient must visit the hospital and request an analysis, but since we have not yet had a first patient case, we cannot officially state that the diagnosis has been implemented," the official explained. "For now, we focus on promoting the DeepDx Prostate, intended for international export, to major hospitals in the country."

The official added that the company remains dedicated to advancing AI technology to improve diagnostic accuracy and efficiency in pathology. The DeepDx Prostate AI algorithm for CNB specimens has shown a sensitivity of 99 percent and a specificity of 97 percent.