Congruence of Deep Learning in Medical Image Processing: Future Prospects and Challenges

Abstract

Machine Learning and deep learning procedures, in particular deep reinforcement neural networks, have quickly become a smart approach for scrutinizing medical signal and image datasets. The perilous problem like Cancer takes place when the cellular reproduction procedure goes out of control when some parts of the human body’s cells arise to spread into contiguous fleshy tissue and divide without discontinuing which may be a clue to death. This fatal cancer is a dangerous disease categorized by undesirable, uncontrolled, and uncoordinated cell division. So early cancer diagnosis can protect a patient. Here, various types of deep learning techniques are applied for optimized feature extraction, normalization or dataset pre-processing (used to eliminate null value, noises, and outlier), data segmentation, accurate classification, and the common description of flow chart is described in Figs. 1 and 7. The survey of deep learning is used for image classification, carotid ultrasound data investigation, cardi tocography, intravascular ultrasound report, lung CT report, brain tumor prediction from the MRI report, object detection, segmentation, breast cancer prediction, ECG (electrocardiogram) signal, EEG (electroencephalogram), PPG signal registration and psoriasis skin disease as well as cancer detection. Concise summaries are delivered of trainings per application zone: pulmonary, musculoskeletal neuro, digital pathology, abdominal, retinal, breast, cardiac. There are various type of deep learning techniques are present to improve accuracy of the medical dataset. Deep reinforcement learning, Recursive neural network, Multilayer perceptron, Recurrent neural network, Boltzmann machine, Convolution neural network are different types of deep learning techniques used to train the image and signal dataset. Generative adversarial network, Auto encoder and deep belief neural network are coming under unsupervised pretrained neural network. Some well known architectural models of convolution neural networks are ResNet (2015), VGGNet (2014), GoogLeNet (2014), ZFNet (2013) is introduced as the visualization concept of the De-convolutional network, AlexNet (2012) and LeNet are basically used to train image dataset, LSTM technique (long short term memory) is used to train signalized dataset and RHSBoost, genetically optimized neural network are used for multiple classification of datasets efficiently. Dimensionality reduction, feature extraction, overfitting, underfitting and normalization problems can be solved using various types of optimization algorithm.