More than 40% of the world’s population suffers from scoliosis. It leads to serious health consequences. The diagnosis of the disease through a routine medical examination or radiological methods is not always effective and safe. Scientists from Perm Polytechnic University have created and trained a neural network to identify key points on the back for diagnosing scoliosis. The use of computer vision will make the disease detection more precise and accessible to patients. The research results are published in the journal «Bulletin of PNIPU. Applied Mathematics and Control Issues.» The work was carried out with the financial support of the Perm Scientific and Educational Center of world level «Rational Subsoil Use.» Scoliosis is particularly common in children, often developing during the period of active growth, starting from the age of five. A healthy spine has physiological curves in the cervical, thoracic, and lumbar regions. In children, the spine is quite flexible, and improper load distribution and other factors lead to deviations of individual vertebrae from the main curve, thus forming scoliosis. Early detection of the disease will help prevent disability, flat feet, circulatory and respiratory disorders, nerve compression, and other complications in children in the future. Diagnosing scoliosis at an early stage is challenging. Currently, it is identified through physical observation by a doctor and radiological methods (X-ray or MRI), which have limitations with frequent repetition. Biometric technologies are currently popular in medicine. They utilize physical and behavioral characteristics of a person and, through computer vision, non-contactly recognize the disease. Scientists from Perm Polytechnic University have developed a project that determines key points on the surface of a person’s back from a photograph using a created neural network algorithm. Polytechnic researchers have been studying new scoliosis detection technologies for several years. Previously, they developed a mathematical algorithm that diagnoses curvature based on a three-dimensional model of the spine. An application interface for phones and its web version are already prepared. Currently, PNIPU scientists have incorporated artificial intelligence into the technology. Together, this allows for a comprehensive assessment of postural abnormalities and musculoskeletal deformities. To train and test the neural network, researchers used 3000 photographs of the backs of adults (18-40 years old) and elementary school students. Key points on all photographs were determined using optical technologies that analyze the body surface image. This allows for remote and non-contact determination of the patient’s torso shape with musculoskeletal system disorders. «We have developed a neural network algorithm that identifies 16 specific points on a photograph of the back. The arrangement of points relative to each other allows conclusions to be drawn about various postural abnormalities. We compared the neural network model with a previously created spatial three-dimensional model based on photogrammetry. With it, a volumetric model can be reconstructed from video footage of the back taken from different angles using a smartphone camera,» shared Vladislav Nikitin, a candidate of physical and mathematical sciences, associate professor of the Department of Computational Mathematics, Mechanics, and Biomechanics at PNIPU. «A doctor or the individual can open the installed program (application) and choose the diagnostic option. Express analysis will determine abnormalities using artificial intelligence based on just one photo, while the extended version will analyze a video file of the back surface captured from different angles. As a result, the person will receive an interpretation of the values and recommendations for preventive exercises,» explained Ivan Shitoev, an assistant at the Department of Computational Mathematics, Mechanics, and Biomechanics at PNIPU. Researchers note that after conducting clinical trials and refining the program, the application will be ready for launch on computers and phones. It can be used by both medical professionals and the general public for scoliosis diagnosis. The development by PNIPU scientists achieves an 85% accuracy rate. The trained neural network can be utilized in clinical medicine, where specialists are interested in the emergence of new and valid tools for diagnosing spinal deformities.
Добавить комментарий