Parallel Multiclass Support Vector Interpretation of Haemodynamic Parameters for Manifestation of Aortic and Arterial Occlusive Diseases

摘要

Aortic and arterial occlusive diseases are congenital conditions manifested in impedance plethysmography and are difficult to interpret. A parallel multiclass support vector classification of haemodynamic parameters computed from plethysmographic observations is proposed for diagnosis of aortoarteritis, atherosclerotic narrowing and coarctation of aorta. The proposed support vector algorithm was able to detect more precisely the presence of thrombotic occlusions at proximal and distal arteries. The proposed method provided better accuracy and sensitivity of 97.46% and 98.3% compared to principal component analysis (PCA) based backpropagation and non-weighted support vector architectures respectively. The results of the genotype were ably supported by receiver operating characteristics (ROC) curves which depict a ratio of true positive rate and false positive rate of over 0.9 for all classes as compared with ratios varying from 0.7 to 0.9 for majority of classes as observed in case of non weighted architecture. A reduction of over 60% in negative likelihood ratio with a 5% increase in negative predictive value was observed as compared to Elman and PCA based backpropagation architectures. The results were validated from angiographic findings at Grant Medical College, J.J. Hospital, and Bhabha Atomic Research Centre (BARC) all in Mumbai. The proposed method also distinguished cases with nephritic syndrome, lymphangitis, and venous disorders against those with arterial occlusive diseases. Application of the proposed method has potential to enhance performance of impedance plethysmography.