LAUSR

Background Osteoporosis is a systemic skeletal disease characterised by a compromised bone resistance that predisposes a subject to an increased risk of fracture. In particular, vertebral fractures are the most common of all osteoporotic fractures. A common clinically-used method for vertebral fracture detection is vertebral morphometry, which is based on measurements of front (Ha), middle (Hm), and rear (Hp) heights of vertebral bodies in radiographic images. This method is quantitative and does not require specific operator skills, but its actual accuracy is affected by errors made during the time-consuming manual measurements. Objectives In this work we propose a fully automatic algorithm for morphometric measurement, whose final goal is to reduce errors due to manual and semi-automatic processes. Our automatic method identifies the vertebrae and their edges. Then, the algorithm measures the characteristic vertebra heights (Ha, Hm, Hp) and determines possible vertebral deformities. Methods The vertebral morphometry uses lateral X-ray images and it is based on height measurements of vertebral bodies: Hp=||P1–P2||2, Ha = ||A1–A2||2, Hm = ||M1–M2||2. Where Pi: rear vertebral corners; Mi: middle vertebral points; Ai: front vertebral corners. The vertebral deformities can be determined as follows: biconcave deformity: Hm/Hp<0.80, crushing deformity: Hp/Hp(±1)<0.80, wedge deformity: Ha/Hp<0.80. The main problem in implementing our fully automatic algorithm was the correct placement of the six reference points for each detected vertebra. Our approach first combined literature-reported methods for the identification of vertebrae in X-Ray images, subsequent emphasise the vertebral borders. The four corners of the vertebra (Pi and Ai) were localised on the borders detected vertebra.While, the middle points Mi were positioned at equal distance from Pi and Ai. Results The performance tests were based on the comparison between the results coming from our automatic approach and those obtained from the manual measurements by an experienced radiologist. We analysed 100 conventional lateral radiographs. The following metrics were used: sensitivity and specificity in vertebra detection; errors in the localization of characteristic points and in the measurement of diagnostic parameters; correlation between manual and automatic measurements. The results of our method showed a sensitivity of 90.8% and a specificity of 99.0%. Average errors in the localisation of vertebral characteristic points were always smaller than 3 mm. Bland-Altman analysis documented a mean error in automatic measurements of diagnostic ratios of 0.01±0.15 (bias ±2 SDs), while Pearson’s correlation coefficient resulted r=0.71 (p<0.001). Conclusions Obtained results of our method compared to the results obtained by a trained radiologist showed an acceptable low error rate, a very good performance in vertebra detection and the same diagnosis (normal, biconcave deformity, crushing deformity, and wedge deformity). Overall, the adopted method has a strong potential for an effective employment in clinical routine for fast and accurate diagnosis of vertebral fractures. Reference [1] Melton LJ, et al. ”Epidemiology of vertebral fractures in women”. in Am J Epidemiol1989May;129(5):1000–1011. Disclosure of Interest None declared