VisionLabs LUNA MEDIUM: recognition of pathologies in medical images

VisionLabs has developed a technology for detecting kidney cancer in non-contrast-enhanced CT scans. The solution is based on pre-trained neural networks that automatically detect and localize regions in the image that differ in density from the surrounding kidney tissue. This will increase the speed of initial image processing and optimize the doctor's workflow. The technology works with non-contrast-enhanced abdominal CT scans. These examinations are conducted in large volumes in clinics and are prescribed even in cases where patients are not suspected of having tumors. Therefore, the doctors working with these images may not have oncology specialization. Additionally, the VisionLabs solution can be particularly useful in regional clinics where there are often no specialists with narrow proficiencies. Image analysis using VisionLabs algorithms helps reduce the burden on doctors and automatically highlights areas that need attention for the final diagnosis. The system does not diagnose on its own and is designed to assist doctors and support medical decision-making. Tasks to be solved: recognition of various pathologies of human internal organs; evaluation of indicators: size and position of formations (cysts/tumors); processing a large volume of medical images, providing initial scoring to relieve the burden on doctors; implementation of automatic retrospective analysis of studies; collection and processing of a large amount of conclusion data The system receives an encrypted file in the standardized DICOM format (Digital Imaging and Communications in Medicine, an industry standard for creating, storing, transmitting, and visualizing digital medical images and documents of examined patients) and extracts the medical image that needs to be processed. It then segments the image into groups of pixels, each representing a meaningful object, and detects malignant formations among them. After processing, the solution provides a response in the required format for external system processing, indicating parameters such as the presence of pathology, its location, size, density, and classification. The image processing and report generation take on average about two to three minutes. Additionally, VisionLabs is currently working on developing new detectors for recognizing liver and adrenal gland pathologies, as well as aortic aneurysms and urolithiasis.