Medical Image Segmentation of Bio-medical Images with Deep Convolution Neural Networks using Ensemble Approach

Image analysis relies heavily on segmentation, which encompasses detecting the features, extraction of features, classification, and treatment. For treatment planning, segmentation assists physicians in quantifying amount of tissue. The segmentation methods are divided into three groups: supervised, unsupervised, and DL segmentation, as well as traditional and computational modelling segmentation. Ensemble techniques in combine many learning models to achieve greater prediction results than each of the individual deep learning algorithms separately. In most scenarios, utilising a very simple ensemble resulted in a lower error rate than using a single model. This demonstrates the efficacy of ensemble approach. This research presents an ensemble deep learning model that was evaluated on ultrasound breast images of women between the ages of 25 and 75 using CNN, MaskR CNN, U Net, and Res Net. Since all four models operate with similar data shape, it makes sensible to design a single input layer that all models will use. The proposed model achieves the highest accuracy of 98.6% on the training, 94.5% validation accuracy, 94.5% test accuracy and a F1 score of 94.32% which outperforms all the individual neural networks.