Literature Review of Deep Learning and Neuroimaging Techniques in Medical Imaging

This literature review discusses the use of deep learning and neuroimaging techniques in medical imaging for diagnosis and treatment of diseases. It covers the application of convolutional neural networks, the NiftyNet platform, and semi- or fully-automated lesion segmentation tools.

Running head: ENGINEERING PROJECT
ENGINEERING PROJECT
Name of the Student:
Name of the University:
Author Note:

1ENGINEERING PROJECT
Literature review of paper 1.
In order to carry out effective diagnosis, detection of the disease in the earlier phases
followed by proper treatment of the diseases, the techniques of magnetic resonance imaging
(MRI), computed tomography (CT), mammography, ultrasound and positron emission
tomography (PET) have been used in the medical sector [1]. Medical images provided with
both physiological and biological data. Earlier these methods were conducted manually by
physicals or other medical personnel, however recent study findings have reported that in
recent times there has been an increase in the analysis by the intervention provided by
computational medical analysis [2]. Another study conducted by [3], has shown that deep
learning methods have been implemented in computational analysis for medical imaging in
order to avoid the hindrances that exists in step of converting engineering into a step of
learning. According to [4], the methods of deep learning can be addressed in situations where
the present samples of the study are quite large. Another study presented deep learning as a
new trend in the field of data analysis, which aims to improve the artificial neural networks.
The study also reflected that the methods of deep learning provides the data predictions in a
multi-layered method [2]. Researches have classified the properties of deep learning into two
major categories namely nonlinear processing units which is of multiple layer nature and
supervised or unsupervised learning of feature presentations. However the early deep learning
frameworks were built on the foundations of artificial neural networks (ANNs). Deep
learning was developed by implementation of a network of layer-wise pre-training deep auto-
encoder (AE). Neural networks are main components behind the deep learning development
[5]. The studies by [1], depicted the application of convolutional neural networks (CNNs) for
the process of medical imaging. The study showed that from mammograms, the ROIs
containing either biopsy-proven masses or normal tissues could be extracted. Certain studies
also showed the application of CNN in the detection of lung nodule [6]. Although in medical

2ENGINEERING PROJECT
applications, deep learning methods have achieved state-of-the-art performance but there are
still some aspects which could be improved. NiftyNet platform is one such improved tool of
implementing deep learning. A high-level deep learning pipeline is given by NiftyNet. It has
the components that are optimized for the purpose of medical imaging like data loading,
sampling and augmentation, networks, loss functions, evaluations, and a model zoo [7].
There are certain interfaces for medical image segmentation that are specific along with the
processes of regression, classification, image generation along with applications of learning
representation.
Literature review of paper 2
Cortical reorganization detection by methods of invasive investigations have been
established as unfeasible methods hence several non-invasive approaches of neuroimaging
were developed like positron-emission tomography (PET) and magnetic resonance imaging
(MRI) [8]. Studies suggested that the neuroimaging studies were involved in the examination
of the cognitive reappraisal. Another study showed that in patients suffering from stroke,
neuroimaging is used to detect enhanced activity in a number of areas such as premotor
cortex, supplementary motor area, parietal and prefrontal cortex along with the striatum,
thalamus and cerebellum during paretic limb movements in both cases of affected and
unaffected hemi- sphere [6]. There has been several meta-analysis studies which stated that
the area of contralesional M1 and the premotor areas that are activated can be found using the
neuroimaging techniques of MRI. There have been studies on the presence of dense intra-
and interhemispheric cortico-cortical projections that exists between the PMC, SMA,
posterior parietal areas and M1 in the brain, can be traced using the above technique [9]. This
helps in facilitating the spinal cord neuron motor outputs. Studies by [10], reflected that

3ENGINEERING PROJECT
experiments conducted related to the longitudinal MRI has successfully shown the existence
of the areas of PMC, SMA and other such regions. The study also revealed that the within the
first days of the commencement of the stroke there has been identification of bilateral activity
that is being detected and after weeks of the concomitant stroke occurrence, motor recovery
has been identified in patients. MRI is primarily carried out for imaging of the brain, in order
to understand the changes that occurs in the brain that occurs in response to a therapeutic
interventions [11]. Stroke neuroimaging studies been seen to utilize for analysis, the semi- or
fully-automated lesion segmentation tools. Some studies have also revealed that fully
automated algorithms depend on the algorithm for identification of the lesion segmentation
[12]. In the studies the Ischemic Stroke Lesion Segmentation (ISLES) challenge is perceived
as an annual satellite challenge. This was discussed in the Medical Image Computing and
Computer Assisted Intervention (MICCAI) meeting. This is based on the provision of a
standardized multimodal clinical MRI dataset which contains about 50–100 brains having
segmented lesions which is done manually [13]. The study mentioned that Anatomical
Tracings of Lesions after Stroke (ATLAS), is a collection of open-source dataset that consists
of 304 T1-weighted MRIs which has segmented diverse lesions done manually and contains
metadata. The prime goal of ATLAS is the provision of a standardized training to the
research community and aims in testing the dataset for lesion segmentation algorithms on the
T1-weighted MRIs [11]. These methods require less amount of human input and expertise
however in spite of this, there persists a requirement of essential computational resources and
time of processing.

4ENGINEERING PROJECT
References
[1] G. Litjens, T. Kooi, B. Bejnordi, A. Setio, F. Ciompi, M. Ghafoorian, J. van der Laak, B.
van Ginneken and C. Sánchez, "A survey on deep learning in medical image analysis",
Medical Image Analysis, vol. 42, pp. 60-88, 2017.
[2] K. Nurzynska, "Deep Learning as a Tool for Automatic Segmentation of Corneal
Endothelium Images", Symmetry, vol. 10, no. 3, p. 60, 2018.
[3] H. Greenspan, B. van Ginneken and R. Summers, "Guest Editorial Deep Learning in
Medical Imaging: Overview and Future Promise of an Exciting New Technique", IEEE
Transactions on Medical Imaging, vol. 35, no. 5, pp. 1153-1159, 2016.
[4] N. Dhungel, G. Carneiro and A. Bradley, "A deep learning approach for the analysis of
masses in mammograms with minimal user intervention", Medical Image Analysis, vol. 37,
pp. 114-128, 2017.
[5] E. Jauch, J. Saver, H. Adams, A. Bruno, J. Connors, B. Demaerschalk, P. Khatri, P.
McMullan, A. Qureshi, K. Rosenfield, P. Scott, D. Summers, D. Wang, M. Wintermark and
H. Yonas, "Guidelines for the Early Management of Patients With Acute Ischemic Stroke: A
Guideline for Healthcare Professionals From the American Heart Association/American
Stroke Association", 2018
[6] A. Rehme and C. Grefkes, "Cerebral network disorders after stroke: evidence from
imaging-based connectivity analyses of active and resting brain states in humans", The
Journal of Physiology, vol. 591, no. 1, pp. 17-31, 2013.

5ENGINEERING PROJECT
[7] E. Gibson, W. Li, C. Sudre, L. Fidon, D. Shakir, G. Wang, Z. Eaton-Rosen, R. Gray, T.
Doel, Y. Hu, T. Whyntie, P. Nachev, M. Modat, D. Barratt, S. Ourselin, M. Cardoso and T.
Vercauteren, "NiftyNet: a deep-learning platform for medical imaging", Computer Methods
and Programs in Biomedicine, vol. 158, pp. 113-122, 2018.
[8] X. Meng, A. Rosenkrantz, N. Mendhiratta, M. Fenstermaker, R. Huang, J. Wysock, M.
Bjurlin, S. Marshall, F. Deng, M. Zhou, J. Melamed, W. Huang, H. Lepor and S. Taneja,
"Relationship Between Prebiopsy Multiparametric Magnetic Resonance Imaging (MRI),
Biopsy Indication, and MRI-ultrasound Fusion–targeted Prostate Biopsy Outcomes",
European Urology, vol. 69, no. 3, pp. 512-517, 2016.
[9] J. Buhle, J. Silvers, T. Wager, R. Lopez, C. Onyemekwu, H. Kober, J. Weber and K.
Ochsner, "Cognitive Reappraisal of Emotion: A Meta-Analysis of Human Neuroimaging
Studies", Cerebral Cortex, vol. 24, no. 11, pp. 2981-2990, 2013.
[10] T. Chen and D. Metaxas, "A hybrid framework for 3D medical image segmentation",
Medical Image Analysis, vol. 9, no. 6, pp. 547-565, 2005.
[11] S. Bauer, R. Wiest, L. Nolte and M. Reyes, "A survey of MRI-based medical image
analysis for brain tumor studies", Physics in Medicine and Biology, vol. 58, no. 13, pp. R97-
R129, 2013.
[12] N. Tajbakhsh, J. Shin, S. Gurudu, R. Hurst, C. Kendall, M. Gotway and J. Liang,
"Convolutional Neural Networks for Medical Image Analysis: Full Training or Fine
Tuning?", IEEE Transactions on Medical Imaging, vol. 35, no. 5, pp. 1299-1312, 2016.
[13] C. Berger, P. Schramm and S. Schwab, "Reduction of Diffusion-Weighted MRI Lesion
Volume After Early Moderate Hypothermia in Ischemic Stroke", Stroke, vol. 36, no. 6, pp.
e56-e58, 2005.

6ENGINEERING PROJECT