A Comparison between Average and Max-Pooling in Convolutional Neural Network for Scoliosis Classification

Abstract

The present study carried out a comparison between average and max-pooling in Convolutional Neural Network for scoliosis classification. In the past, around 2 to 4 per cent of adolescence has been reported to suffer with scoliosis. Currently, radiographic is the clinical approach used in identifying the Cobb angle to determine the suitable treatment for this category of patients. However, over exposure to radiographic have been seen to what is leading to the risk of cancer. As such, the present study proposed the used of photogrammetric approach to overcome the radiographic side effect. The photogrammetric of human’s back is acquired to classify the scoliosis into Lenke Type 1 or Non-Type 1. Due to limited dataset, rotation, x-transition and y-transition of data augmentation was carried out. These data are classified using convolutional neural network. The convolutional neural network (CNN) consist of convolve layer, pooling layer, fully connected layer and softmax layer. Selection of the best pooling layer is important to increase the accuracy of classification. As mentioned earlier, the present study compares between average and max-pooling layer to classify the Lenke classification system. The result shows that the use of max-pooling can achieve a higher accuracy which is 84.6% compared to average pooling. Future studies are encouraged to collect more data to further prove the effectiveness of max-pooling layer.