AI-Driven Software Based Off NIH Algorithm Receives FDA Clearance in Detection and Diagnosis of Prostate Cancer
An Omaha-based MRI medical device company, Bot Image, Inc., received FDA clearance for its artificial intelligence (AI) software used to improve the accuracy and speed of prostate cancer detection. The tool, called ProstatID, combines AI with traditional MRI scanning. The conceptual AI algorithm was first developed by scientists and physicians at NIH. Its algorithm recognizes and measures the volume of the prostate gland and detects suspicious cancerous lesions. The software can assess visible features such as brightness and grayscale, but also those invisible to the human eye such as pixel-by-pixel texture, 3D voxel texture, and infinite feature weighting testing and validation.
Peter L. Choyke, M.D., F.A.C.R., chief of the Molecular Imaging Branch at NCI’s Center for Cancer Research, and Baris Turkbey, M.D., F.S.A.R., senior clinician of the Molecular Imaging Branch, contributed to the 2017 study, “Detection of prostate cancer in multiparametric MRI using random forest with instance weighting.”
“MRI of the prostate has become an important diagnostic method for prostate cancer. It can localize areas within the prostate that might harbor cancer. These areas can then be directly biopsied using MRI-Ultrasound fusion biopsy, a method that was also developed in the Molecular Imaging Branch,” explains Dr. Choyke. “We began thinking about using AI to train computers to read MRIs as well as we do at NIH. Immediate benefits were that the readings were more consistent and that relatively inexperienced readers began to perform almost as well as more experienced readers.”
Dr. Turkbey adds, "for this technique to be successful, prostate MRI studies need to be read accurately. This includes key steps such as detection of cancer-suspicious intraprostatic lesions and assigning correct risk category for predicting presence of clinically significant prostate cancer.”
Bot Image projects that annual MRI screening with AI could become a new screening method for prostate cancer. They anticipate a 5-step process:
- The physician will order a non-invasive prostate MRI for the patient.
- The MRI scans will be acquired rapidly.
- These images will be uploaded to ProstatID’s cloud-hosted platform.
- ProstatID’s algorithm will segment the prostate and generate a probability map of cancer.
- Radiologists will review the map and send the final results to the physician.
Dr. Choyke adds, “the radiologist is still free to interpret the MRI the way he or she sees fit but is also free to incorporate the AI’s ‘opinion’ into the interpretation. This should help make interpretations of MRIs more consistent and more reliable by decreasing the human element.”
This has been a project years in the making. ScanMed, LLC, acquired the rights to NIH’s AI prototype in March of 2018. Five months later, Bot Image was officially formed. Dr. Turkbey remembers this milestone and says, “based on the results of our studies, [ScanMed] showed an interest for the AI model and proceeded with the licensing process. After completing the required regulatory steps, this AI model got the FDA’s approval.”
“We are very happy to see that our work is finding its way into FDA-approved products that will help patients with prostate cancer get quicker and more accurate diagnoses,” Dr. Choyke contributes.
Prostate cancer is the most common cancer and the second leading cause of cancer death among men in the United States.