Stats, burdens, and innovation in the field
In 2015, nearly 575’00 MRI exams were performed in Switzerland (>66/1’000 inhabitants) across 181 MRI machines distributed over 288 hospitals and specialized clinics. This analyses were all performed by radiologists, which accounted to less than 2.5% in the Swiss medical community. Although MRI rank as the main field of interest for over 83% of radiologists in Switzerland, an average of circa 700 MRIs per year (2.7 MRI/day) for each radiologist is likely to tire anybody. If we consider an average of 570 images/examination, this means that each radiologist should go through 2.6 images/minute each single day, supposing they would all work full-time and solely on MRI. In fact, I am not even mentioning all the other imaging techniques radiologists need to examine, like CT and PET scans. Neither am I considering all the other duties in a typical day of a radiologist. In practice, radiologists need to look at the MRI much faster than this theoretical value.
Let me make this straight: I’m not an MD. But I experienced the pain of going through hundreds of slices from 3D X-ray CT scans. Hundreds of images just for one single samples. It is terribly boring. What’s even more frustrating is knowing that acquiring the images doesn’t take too long. The machine works fast. It’s the human component that is “slow”, if 2.6 images/minute can be considered that. Anything that would make the process faster and less repetitive would be beneficial. However, as humans, there’s not much we can do. Interpreting correctly the meaning of the different features visible in an MRI, and providing a correct diagnosis is not an easy task. Therefore, hurrying up while going through the many MRI slices is riskier: features may be missed, increasing the chances of misdiagnosis. According to a research from Johns Hopkins University, in general “the retrospective error rate among radiologic examinations is approximately 30%, with real-time errors in daily radiology practice averaging 3–5%”, imagine what would happen increasing the speed. Are we hopeless?
Balzano Informatics doesn’t think so. We believe that cognitive computing can revolutionize the way MR images are interpreted. Our goal is to create a solution that can be integrated into the existing radiology practice, so it can become a seamless part of routine care, reducing the cost or burden on the healthcare system, reducing the workload for radiologists, and minimizing the risk of misdiagnosis.
With over 20% MRIs performed in Switzerland focusing on joints, particularly knees, it appears clear why ScanDiags, Balzano’s solution, begins this automation process tackling one of the most common knee injury: Meniscal tear. The artificial neural network developed is trained on medical images, and holds a vast domain-specific knowledge to identify structures and abnormalities in knee MRIs. It provides a report for the doctor in a matter of seconds. In this way, radiologist can get a second opinion, interpret an MRI, and develop a diagnosis within minutes.