ABSTRACT
Neuromyelitis optica spectrum disorder (NMOSD), also known as Devic disease, is an autoimmune central nervous system disorder in humans that commonly causes inflammatory demyelination in the optic nerves and spinal cord. Inflammation in the optic nerves is termed optic neuritis (ON). ON is a common clinical presentation; however, it is not necessarily present in all NMOSD patients. ON in NMOSD can be relapsing and result in severe vision loss. To the best of our knowledge, no study utilises deep learning to classify ON changes on MRI among patients with NMOSD. Therefore, this study aims to deploy eight state-of-the-art CNN models (Inception-v3, Inception-ResNet-v2, ResNet-101, Xception, ShuffleNet, DenseNet-201, MobileNet-v2, and EfficientNet-B0) with transfer learning to classify NMOSD patients with and without chronic ON using optic nerve magnetic resonance imaging. This study also investigated the effects of data augmentation before and after dataset splitting on cropped and whole images. Both quantitative and qualitative assessments (with Grad-Cam) were used to evaluate the performances of the CNN models. The Inception-v3 was identified as the best CNN model for classifying ON among NMOSD patients, with accuracy of 99.5%, sensitivity of 98.9%, specificity of 93.0%, precision of 100%, NPV of 99.0%, and F1-score of 99.4%. This study also demonstrated that the application of augmentation after dataset splitting could avoid information leaking into the testing datasets, hence producing more realistic and reliable results.
ABSTRACT
20% of brain tumor patients present with seizures at the onset of diagnosis, while a further 25-40% develop epileptic seizures as the tumor progresses. Tumor-related epilepsy (TRE) is a condition in which the tumor causes recurring, unprovoked seizures. The occurrence of TRE differs between patients, along with the effectiveness of treatment methods. Therefore, determining the tumor properties that correlate with epilepsy can help guide TRE treatment. This article reviews the MRI sequences and image post-processing algorithms in the study of TRE. It focuses on epilepsy caused by glioma tumors because it is the most common type of malignant brain tumor and it has a high prevalence of epilepsy. In correlational TRE studies, conventional MRI sequences and diffusion-weighted MRI (DWI) are used to extract variables related to the tumor radiological characteristics, called imaging factors. Image post-processing is used to correlate the imaging factors with the incidence of epilepsy. The earlier studies of TRE used univariate and multivariate analysis to study the correlations between specific variables and incidence of epilepsy. Later, studies used voxel-based morphometry and voxel lesion-symptom mapping. Radiomics has been recently used to post-process the images for the study of TRE. This article will discuss the limitation of the existing imaging modalities and post-processing algorithms. It ends with some suggestions and challenges for future TRE studies.
ABSTRACT
Coronavirus disease 2019 (COVID-19) is a disease caused by a novel strain of coronavirus, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), severely affecting the lungs. Our study aims to combine both quantitative and qualitative analysis of the convolutional neural network (CNN) model to diagnose COVID-19 on chest X-ray (CXR) images. We investigated 18 state-of-the-art CNN models with transfer learning, which include AlexNet, DarkNet-19, DarkNet-53, DenseNet-201, GoogLeNet, Inception-ResNet-v2, Inception-v3, MobileNet-v2, NasNet-Large, NasNet-Mobile, ResNet-18, ResNet-50, ResNet-101, ShuffleNet, SqueezeNet, VGG-16, VGG-19, and Xception. Their performances were evaluated quantitatively using six assessment metrics: specificity, sensitivity, precision, negative predictive value (NPV), accuracy, and F1-score. The top four models with accuracy higher than 90% are VGG-16, ResNet-101, VGG-19, and SqueezeNet. The accuracy of these top four models is between 90.7% and 94.3%; the F1-score is between 90.8% and 94.3%. The VGG-16 scored the highest accuracy of 94.3% and F1-score of 94.3%. The majority voting with all the 18 CNN models and top 4 models produced an accuracy of 93.0% and 94.0%, respectively. The top four and bottom three models were chosen for the qualitative analysis. A gradient-weighted class activation mapping (Grad-CAM) was used to visualize the significant region of activation for the decision-making of image classification. Two certified radiologists performed blinded subjective voting on the Grad-CAM images in comparison with their diagnosis. The qualitative analysis showed that SqueezeNet is the closest model to the diagnosis of two certified radiologists. It demonstrated a competitively good accuracy of 90.7% and F1-score of 90.8% with 111 times fewer parameters and 7.7 times faster than VGG-16. Therefore, this study recommends both VGG-16 and SqueezeNet as additional tools for the diagnosis of COVID-19.
ABSTRACT
The optic nerve is known to be one of the largest nerve bundles in the human central nervous system. There have been many studies of optic nerve imaging and post-processing that have provided insights into pathophysiology of optic neuritis related to multiple sclerosis and neuromyelitis optica spectrum disorder, glaucoma, and Leber's hereditary optic neuropathy. There are many challenges in optic nerve imaging, due to the morphology of the nerve through its course to the optic chiasm, its mobility due to eye movements and the high signal from cerebrospinal fluid and orbital fat surrounding the optic nerve. Recently, many advanced and fast imaging sequences have been used with post-processing techniques in attempts to produce higher resolution images of the optic nerve for evaluating various diseases. Magnetic resonance imaging (MRI) is one of the most common imaging methodologies for the optic nerve. This review paper will focus on recent MRI advances in optic nerve imaging and explain several post-processing techniques being used for analysis of optic nerve images. Finally, some challenges and potential for future optic nerve studies will be discussed.
Subject(s)
Multiple Sclerosis , Optic Neuritis , Humans , Magnetic Resonance Imaging , Multiple Sclerosis/diagnostic imaging , Optic Chiasm , Optic Nerve/diagnostic imaging , Optic Neuritis/diagnostic imagingABSTRACT
An effective and practical Image Quality Assessment (IQA) model is needed to assess the image quality produced from any new hardware or software in MRI. A highly competitive No Reference - IQA (NR - IQA) model called Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) initially designed for natural images were modified to evaluate structural MR images. The BRISQUE model measures the image quality by using the locally normalized luminance coefficients, which were used to calculate the image features. The modified-BRISQUE model trained a new regression model using MR image features and Difference Mean Opinion Score (DMOS) from 775 MR images. Two types of benchmarks: objective and subjective assessments were used as performance evaluators for both original and modified-BRISQUE models. There was a high correlation between the modified-BRISQUE with both benchmarks, and they were higher than those for the original BRISQUE. There was a significant percentage improvement in their correlation values. The modified-BRISQUE was statistically better than the original BRISQUE. The modified-BRISQUE model can accurately measure the image quality of MR images. It is a practical NR-IQA model for MR images without using reference images.
Subject(s)
Image Processing, Computer-Assisted , Magnetic Resonance Imaging , Adult , Algorithms , Female , Humans , Models, Statistical , Normal Distribution , Observer Variation , Radiography , Regression Analysis , Software , Support Vector Machine , Young AdultABSTRACT
Medical Image Quality Assessment (IQA) plays an important role in assisting and evaluating the development of any new hardware, imaging sequences, pre-processing or post-processing algorithms. We have performed a quantitative analysis of the correlation between subjective and objective Full Reference - IQA (FR-IQA) on Magnetic Resonance (MR) images of the human brain, spine, knee and abdomen. We have created a MR image database that consists of 25 original reference images and 750 distorted images. The reference images were distorted with six types of distortions: Rician Noise, Gaussian White Noise, Gaussian Blur, DCT compression, JPEG compression and JPEG2000 compression, at various levels of distortion. Twenty eight subjects were chosen to evaluate the images resulting in a total of 21,700 human evaluations. The raw scores were then converted to Difference Mean Opinion Score (DMOS). Thirteen objective FR-IQA metrics were used to determine the validity of the subjective DMOS. The results indicate a high correlation between the subjective and objective assessment of the MR images. The Noise Quality Measurement (NQM) has the highest correlation with DMOS, where the mean Pearson Linear Correlation Coefficient (PLCC) and Spearman Rank Order Correlation Coefficient (SROCC) are 0.936 and 0.938 respectively. The Universal Quality Index (UQI) has the lowest correlation with DMOS, where the mean PLCC and SROCC are 0.807 and 0.815 respectively. Student's T-test was used to find the difference in performance of FR-IQA across different types of distortion. The superior IQAs tested statistically are UQI for Rician noise images, Visual Information Fidelity (VIF) for Gaussian blur images, NQM for both DCT and JPEG compressed images, Peak Signal-to-Noise Ratio (PSNR) for JPEG2000 compressed images.
Subject(s)
Abdomen/diagnostic imaging , Brain/diagnostic imaging , Image Processing, Computer-Assisted/methods , Knee/diagnostic imaging , Magnetic Resonance Imaging/methods , Spine/diagnostic imaging , Adult , Algorithms , Artifacts , Databases, Factual/statistics & numerical data , Female , Humans , Magnetic Resonance Imaging/statistics & numerical data , Male , Normal Distribution , Signal-To-Noise Ratio , Young AdultABSTRACT
Direct-MR neuronal detection (DND) of transient magnetic fields has recently been investigated as a novel imaging alternative to the conventional BOLD functional MRI (fMRI) technique. However, there remain controversial issues and debate surrounding this methodology, and this study attempts clarification by comparing BOLD responses in the human visual system with those of DND. BOLD relies on indirectly measuring blood oxygenation and flow changes as a result of neuronal activity, whereas the putative DND method is based on the hypothesis that the components of the in vivo neuronal magnetic fields, which lie parallel to the B(0) field, can potentially modulate the MR signal, thus providing a means of direct detection of nerve impulses. Block paradigms of checkerboard patterns were used for visual stimulation in both DND and BOLD experiments, allowing detection based on different frequency responses. This study shows colocalization of some voxels with slow BOLD responses and putative fast DND responses using General Linear Model (GLM) analysis. Frequency spectra for the activated voxel cluster are also shown for both stimulated and control data. The mean percentage signal change for the DND responses is 0.2%, corresponding to a predicted neuronal field of 0.14 nT.
Subject(s)
Brain Mapping/methods , Evoked Potentials, Visual/physiology , Image Enhancement/methods , Image Interpretation, Computer-Assisted/methods , Magnetic Resonance Imaging/methods , Neurons/physiology , Visual Cortex/physiology , Adult , Algorithms , Female , Humans , Male , Middle Aged , Neurons/cytology , Reproducibility of Results , Sensitivity and Specificity , Visual Cortex/cytologyABSTRACT
PURPOSE: To investigate the possibility of detecting visually-evoked axonal currents in the splenium of the human corpus callosum using a 3.0T MRI system. MATERIALS AND METHODS: Axonal currents produce weak and transient magnetic fields, and the components of these that lie parallel to the B(0) field of the MRI system can potentially modulate the MR signal, which can be detected as an integrated effect over time. A fast gradient-echo echo-planar imaging (GE-EPI) sequence with short TR and intermediate TE was employed in an attempt to detect such axonal currents using light-emitting diode (LED) visual stimulation paradigms. RESULTS: The mean magnitude signal change, expressed relative to the fully relaxed equilibrium signal calculated from the measured value using the known T1 of white matter, was 0.014 +/- 0.004% at TE = 30 msec. This corresponded to a mean axonal field of 0.11 +/- 0.03 nT, according to the hypothesis that the axonal currents create a Lorentzian field distribution within an imaging voxel. CONCLUSION: Measured frequency spectra and statistical mapping using the general linear model (GLM) showed evidence of the stimulus localized within the splenium of the corpus callosum, which was not thought to be due to motion artifacts or physiological responses.
Subject(s)
Axons/pathology , Corpus Callosum/anatomy & histology , Corpus Callosum/pathology , Echo-Planar Imaging/methods , Magnetic Resonance Imaging/methods , Action Potentials , Adult , Axons/metabolism , Electromagnetic Fields , Female , Humans , Image Processing, Computer-Assisted , Light , Male , Middle Aged , Models, Statistical , Photic StimulationABSTRACT
Neuronal currents produce weak transient magnetic fields, and the hypothesis being investigated here is that the components of these parallel to the B0 field can potentially modulate the MR signal, thus providing a means of direct detection of nerve impulses. A theory for the phase and amplitude changes of the MR signal over time due to an external magnetic field has been developed to predict this modulation. Experimentally, a fast gradient-echo EPI sequence (TR = 158 ms, TE = 32.4 ms) was employed in an attempt to directly detect these neuronal currents in the adult human optic nerve and visual cortex using a 280-mm quadrature head coil at 1.5 T. A symmetrical intravoxel field distribution, which can be plausibly hypothesized for the axonal fields in the optic nerve and visual cortex, would result in phase cancellation within a voxel, and hence, only amplitude changes would be expected. On the other hand, an asymmetrical intravoxel field distribution would produce both phase and amplitude changes. The in vivo magnitude image data sets show a significant nerve firing detection rate of 56%, with zero detection using the phase image data sets. The percentage magnitude signal changes relative to the fully relaxed equilibrium signal fall within a predicted RMS field range of 1.2-2.1 nT in the optic nerve and 0.4-0.6 nT in the visual cortex, according to the hypothesis that the axonal fields create a symmetrical Lorentzian field distribution within the voxel.
Subject(s)
Action Potentials/physiology , Axons/physiology , Magnetic Resonance Imaging/methods , Optic Nerve/physiology , Visual Cortex/physiology , Adult , Computer Simulation , Electromagnetic Fields , Female , Humans , Image Processing, Computer-Assisted , Male , Phantoms, Imaging , Signal Processing, Computer-AssistedABSTRACT
The aim of this study was to directly detect spectral components of the magnetic fields of ionic currents caused by firing of the axons in the optic nerve in response to visual strobe stimulation. The magnetic field parallel to the main B0 field can potentially alter the local phase and magnitude of the MR signal which can cause signal loss due to intravoxel dephasing. Measured frequency spectra showed evidence of the strobe stimulus localized to regions containing the optic nerve, not thought to be due to motion artifacts, in 30 out of 52 experiments in 5 adult human subjects. The effect was (0.15 +/- 0.05)% of the mean magnitude equilibrium signal from the voxel in the frequency range 0.7-3.3 Hz, corresponding to an estimated field of (1.2 +/- 0.4) nT, at an echo time of TE = 32.4 ms using a 1.5 T MRI scanner. Only 1 of 12 phase image experiments showed effects. These findings provide preliminary evidence for direct detection of axonal firing in the optic nerve.