Developing lightweight cnn models for detecting lesions in the medical imaging

Abstract

Classification and characterization of the lesions in medical imaging are getting interested newlinein the field of computer-aided diagnostics (CAD) systems for radiology. A num newlineber of machine learning (ML) algorithms are proposed for classification and newlinecharacterization. All the existing models suffer from inter-observer variability. CNN newlinemodels are based on Deep Neural Networks, which are distinct from typical Machine newlineLearning methods such as k-NN, Decision-Trees, and so on. Moreover, the performance newlineof Deep Neural Network-based approaches is better as compared to traditional Machine newlineLearning approaches as these models extract the features from training data automatically. newlineIn the past, the researchers have proposed many CNN architectures such as VGG16, newlineVGG19, Inception V3, MobileNet, ResNet50, etc. for the classification of 1000 class newlineImageNet datasets. These models can also be utilized for the classification of other datasets newlineby transfer learning. While pre-trained CNN models work well, the high number of newlineparameters required makes them computationally intensive. newlineIn this thesis, we proposed several simplified and optimized CNN models for the newlineclassification and characterization of the lesions in the carotid artery, lung, and brain newline(Wilson disease). The datasets are taken from the medical practitioners at different newlinegeological locations and several publicly available datasets. We got the ground truth from newlinethe radiologist for real-time data. Then we augmented the data into folds. In these cohorts, newlinewe ran all of the recommended models, including CNN models, Transfer Learning (TL) newlinemodels, and Hybrid CNN models. We achieved an accuracy of 95.66% in carotid using newlineoptimized CNN and 99.45% accuracy using hybrid CNN, 98.9% in lung, and 91.2% in newlineWilson disease. We validated our models and hypothesis on the characterization of the newlinemulticenter study. We compared the performance of the AI models on various performance metrices and then ranked the models based on the grading schema for the best classifier.