The hybrid feature extraction method for classification of adolescence idiopathic scoliosis using Evolving Spiking Neural Network

Abstract

Scoliosis is the most commonly occurring spinal condition in adolescents. Photogrammetry is an alternative to radiography for examining adolescents with idiopathic scoliosis (AIS) to measure the curvature of the human back and avoid radiography exposure. Currently, a manual approach is applied to detect the spine curve. Machine Learning (ML) models are introduced to overcome this issue and reduce human error. Thus, an appropriate Features Extraction (FE) method is crucial to producing a good ML classification model. Local Binary Pattern (LBP) is capable of providing the desired features from the surface of a human body. However, it produces a high number of parameters for each image, affecting the classification performance. Therefore, a new FE method has been proposed to reduced the number of parameters. A fusion of LBP and 1DCNN (1-Dimensional Convolutional Neural Network) FE methods that can reduce the number of parameters by 50% is proposed. These features are classified using Evolving Spiking Neural Network (LBP-1DCNN-ESNN) and achieve an accuracy of 90.00%. It leads to faster and more accurate results. ESNN can produce efficient and fast decisions through a one-pass learning process generated from the neuron repository. The proposed LBP-1DCNN-ESNN model benefits healthcare when quick solutions are demanded.

Keywords

• Adolescent idiopathic scoliosis
• Local binary pattern
• Convolution neural network
• Evolving Spiking Neural Network
• Photogrammetry