Efficiency of the decision support system for determining the severity of scoliotic spinal deformity by the analysis of radiographs

Abstract

BACKGROUND: In 2020, a decision support system was developed at expense of grant from Innovation Promotion Fund and integrated into medical base of prosthetic and orthopedic center “Sсoliologic.ru.” It is the first computer program for determining severity of scoliotic spinal deformity in analysis of radiographs. AIM: To evaluate performance of a computer program for determining Cobb angle by comparing obtained automated data with manual measurement data by various specialists. MATERIALS AND METHODS: From medical base of prosthetic and orthopedic center “Sсoliologic.ru,” 411 digital radiographs of spine of children and adolescents were selected, which were drawn by a radiologist (BP-etalon), an orthopedic traumatologist with 5 years of experience and experience in vertebrology (OO), and an orthopedic traumatologist with less than 1 year of experience and no experience in vertebrology (OB) and computer program (KP). The number and proportion of exact matches and mean absolute percentage error (MAPE) of results of measuring the Cobb angle of reference data were compared with indicators obtained by OO, OB, and KP in, for various types of scoliosis according to the classification of Rigо, as well as in determining the main arc. Bland–Altman graphs were used. The mean absolute deviation (MAD) and MAPE for the main arc with magnitude ranging from 20° to 50° were calculated. RESULTS: The exact matches of KP with standard were 21%. Hence, every fifth measurement of the Cobb angle on X-ray, calculated automatically, was identical to standard. The highest proportion of matches with standard was noted in OO (33%), and the lowest in OB (16%). The highest accuracy of KP was found at types E1 and E2, and the arc of 3 was 30% and 24%, respectively. The proportion of KP matches with standard increased to 61%, with an error of 3°, and to 75%, with an error of 5°; with an error of 7°, the proportion of matches reached 84%. Moreover, the proportion of coincidences of a non-experienced researcher increased from 34% to 58%. When determining magnitude of the main arc of scoliosis, OO increased his/her accuracy by almost 2,6 times, KP by 2,1 times, and OB by 1,5 times. The MAD of KP for all major arcs was 3.8°. Good reliability of the obtained KP indicators compared to the reference (ICC = 0.9) was noted. CONCLUSIONS: The use of a computer program may be recommended for automated determination of Cobb angle, because current algorithm of computer program has been found to be accurate compared to reference measurement. Its accuracy was significantly higher than that of novice orthopedic traumatologists with no experience in vertebrology.