Ensemble Pre-Trained Deep Convolutional Neural Network Model for Classifying Medical Image Datasets

Abstract

Over the last few years, Deep Learning models have shown prominent results in medical image analysis especially to predict disease at the earlier stages. Since Deep Neural Network require more training data for better prediction, it needs more computational time for training. Transfer learning is a technique which uses the learned knowledge to perform the classification task by minimizing the number of training data and training time. To increase the accuracy of a single classifier, ensemble learning is used as a meta-learner. This research work implements a framework Ensemble Pre-Trained Deep Convolutional Neural Network using Resnet50, InceptionV3 and VGG19 pre-trained Convolutional Neural Network models with modified top layers to classify the disease present in the medical image datasets such as Covid X-Rays, Covid CT scans and Brain MRI with less computational time. Further, these models are combined using stacking and bagging ensemble approach to increase the accuracy of single classifier. The datasets are distributed as train, test and validation data and the models are trained and tested for four epochs. All the models are evaluated using validation data and the result shows that the ensemble learning approach increases the prediction accuracy when compared to the single models for all the datasets. In addition, this experiment reveals that the stacked model attains higher test accuracy of 99% for chest X-Ray images, 100% for chest CT scan images and 98% for brain MRI, compared to the bagged models.