Integrating Weighted Average with Deep Learning Methods for Neuroblastoma Detection Using Computed Tomography Images

Abstract

Neuroblastoma needs to be detected in its early stages, otherwise, it gets worse and cannot be treated once it spreads to other parts of the body. The survival rate of neuroblastoma patients can be increased by adopting the right diagnosis. Recently, pre-trained models have become popular for the detection of cancer from Computed Tomography (CT) images. The detection of neuroblastoma is addressed using CT scans with the help of pre-trained Convolution Neural Network models which consumes less training time. The research is based on two different datasets of CT scan images which are taken from Mayo Hospital Lahore and an open-source platform. Further, a labeled dataset is used for investigating results that are based on CT images of neuroblastoma. In this research, the proposed technique is based on the combination of VGG19, VGG 16, ResNet 50, inception V3, and weighted average ensemble that can detect neuroblastoma with higher accuracy and also validated on two different datasets. Each of the models has given some weight by implementing the method of weighted average ensemble. These pre-trained models are evaluated through evaluation matrices namely, recall, precision, support, and f-1 score are used to evaluate the performance of the proposed framework. The VGG19 model reported higher accuracies for Dataset 1 and Dataset 2, 99.80% and 98.70% respectively.