Diagnostic Categorization and Neurocognitive Prediction Employing Neuroimaging Data Using Deep Learning in Alzheimer's Illness

Diagnostic Categorization and Neurocognitive Prediction Employing Neuroimaging Data Using Deep Learning in Alzheimer's Illness

Srividya Bharadwaja, Smitha Sasi
Copyright: © 2022 |Pages: 22
DOI: 10.4018/978-1-7998-9534-3.ch013
OnDemand:
(Individual Chapters)
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Traditional analytic strategies for investigating neuroimaging biomarkers for neuropsychiatric illnesses have relied on mass univariate statistics, assuming that various brain areas function separately. Machine learning (ML) methods that take into account intercorrelation across areas have recently become a popular and important part of computer-assisted analytical procedures and are now frequently used for the automated diagnosis and analysis of neuropsychiatric illnesses. The goal of this chapter is to provide a detailed overview of CNN and RNN applications in medical image comprehension. The overarching goal is to encourage medical image understanding experts to use CNNs extensively in their research and diagnosis. This chapter describes the development of various novel DL-based approaches and models as well as advancements in high-speed computing techniques, which provide a once-in-a-lifetime chance to anticipate and control Alzheimer's disease.
Chapter Preview
Top

Overview Of Deep Learning Techniques

Deep learning is a type of machine learning that learns features through a hierarchical learning process. Methods of deep learning for a variety of domains, categorization and prediction have been used. Including natural language processing and computer vision both of which are show performance breakthroughs. DL is a branch of machine learning that can be used to create models that extract high-dimensional characteristics from data. It has gotten a lot of attention in recent years, notably in the field of image analysis. A number of deep learning architectures have been published in the literature, including CNN, DNN, RNN, AE, Deep Belief Network (DBN), and Probabilistic Neural Network (PNN) (Gulshan, 2016).

Convolutional Neural Network (CNN):

A CNN, also known as a ConvNet, uses learnable weights and biases to apply to distinct regions of an input image, allowing one image to be distinguished from another.

CNN employs convolution instead of basic matrix multiplication in at least one of their layers. It's most commonly used in unstructured datasets. 2D-CNN predicts segmentation maps for a single slice using 2D-convolutional kernels.

Only spatial dimensions such as height and width can be used by 2D-CNN. Context information from adjacent slices cannot be recovered since 2D-CNN only accepts one slice as input. In terms of utility, voxel data from neighboring slices may be sufficient for categorization tasks. By predicting the volumetric patch of neuroimaging data, 3D-CNN, on the other hand, can retain temporal dimensions. Although the capacity of 3D-CNNs to anchor interslice context information improves performance, it comes at a cost in terms of computation time and the number of parameters that 3D-CNNs must utilize. Figure 1 shows CNN Architecture

Figure 1.

CNN architecture

978-1-7998-9534-3.ch013.f01

Complete Chapter List

Search this Book:
Reset