Assessment of axial postural abnormalities in Parkinsonism: automatic picture analysis software

BackgroundSoftware-based measurements of axial postural abnormalities in Parkinson's disease (PD) are the gold standard but may be time-consuming and not always feasible in clinical practice. An automatic and reliable software to accurately obtain real-time spine flexion angles according to the recently proposed consensus-based criteria would be a useful tool for both research and clinical practice. ObjectiveWe aimed to develop and validate a new software based on Deep Neural Networks to perform automatic measures of PD axial postural abnormalities. MethodsA total of 76 pictures from 55 PD patients with different degrees of anterior and lateral trunk flexion were used for the development and pilot validation of a new software called AutoPosturePD (APP); postural abnormalities were measured in lateral and posterior view using the freeware NeuroPostureApp (gold standard) and compared with the automatic measurement provided by the APP. Sensitivity and specificity for the diagnosis of camptocormia and Pisa syndrome were assessed. ResultsWe found an excellent agreement between the new APP and the gold standard for lateral trunk flexion (intraclass correlation coefficient [ICC] 0.960, IC95% 0.913-0.982, P < 0.001), anterior trunk flexion with thoracic fulcrum (ICC 0.929, IC95% 0.846-0.968, P < 0.001) and anterior trunk flexion with lumbar fulcrum (ICC 0.991, IC95% 0.962-0.997, P < 0.001). Sensitivity and specificity were 100% and 100% for detecting Pisa syndrome, 100% and 95.5% for camptocormia with thoracic fulcrum, 100% and 80.9% for camptocormia with lumbar fulcrum. ConclusionsAutoPosturePD is a valid tool for spine flexion measurement in PD, accurately supporting the diagnosis of Pisa syndrome and camptocormia.