MRI-based Classification Research Articles

ALSA-3: Customized CNN model through ablation study for Alzheimer's disease classification

Alzheimer's disease (AD), a prevalent neurological condition, poses a multifaceted challenge affecting millions worldwide. It demands diverse solutions, both pharmaceutical and non-pharmaceutical, to ameliorate symptoms and enhance patients' quality of life across various stages. Tailored intervention strategies are essential for addressing AD's progression, emphasizing the significance of early detection and stage classification. This study underscores the pivotal role of medical imaging, specifically MRI, in AD diagnosis and monitoring. MRI interpretation is often time-consuming and relies on clinical intuition, necessitating more efficient and accurate approaches. In response, a custom convolutional neural network (CNN) model named ALSA-3 and deep learning (DL) techniques are explored for MRI-based alzheimer's disease classification. Various image processing techniques, such as MSE, RMSE, PSNR, and SSIM, enhance image quality. An ablation study systematically adjusts hyperparameters and layer structures. Results showcase the superiority of the proposed ALSA-3 model over traditional methods, achieving an exceptional 99.50% accuracy, with precision, F1-score, and recall values of 100%, 99%, and 99%. Model robustness is evaluated with different k values, reinforcing its efficacy in alzheimer's disease classification. Moreover, the elucidation of the ALSA-3 model incorporates grad-cam and grad-cam++ explainable artificial intelligence (XAI) techniques, to furnish comprehensive explanations elucidating the clarify the reasoning behind model decisions. This study not only advances state-of-the-art in alzheimer's disease classification but also holds the potential to significantly benefit patients, caregivers, and the medical community. The improved accuracy and efficiency of our model promise more accurate and timely diagnosis, thereby enhancing the lives of those affected by alzheimer's disease.

An isolated pontine infarct extending to the basal pontine surface has a higher abnormal ankle-brachial index

Patients with ischemic stroke and branch atheromatous disease (BAD) have worse neurological deficits and prognoses than those with small vessel occlusion (SVO). However, both disorders are forms of deep brain infarctions. This study aimed to investigate an MRI-based etiological classification for isolated pontine infarctions and assess differences in vascular risk factors and peripheral arterial disease among etiological subtypes. Consecutive data of patients admitted for acute ischemic stroke or transient ischemic attack between August 2016 and July 2019 were reviewed. Acute isolated pontine infarcts were classified into 3 groups: BAD, SVO, and large-artery atherosclerosis (LAA), according to basilar or vertebral artery steno-occlusion and the extent of the infarct lesion on the basal pontine surface as displayed on magnetic resonance imaging and angiography. Vascular risk factors, ankle-brachial index (ABI), and brachial-ankle pulse wave velocity were analyzed in the 3 groups. Among 64 enrolled patients, BAD was the most common cause of isolated pontine infarct. The BAD group had a higher frequency of abnormal ABI and hypertension than the SVO group. The BAD group had abnormal ABI and hyperlipidemia more frequently than the LAA group. No significant difference was found in diabetes or brachial-ankle pulse wave velocity incidence between the BAD and SVO groups. ABI and vascular risk factors in the BAD group were more similar to those in the LAA group than to those in the SVD group. This finding suggests that pontine lesions extending to the basal pontine surface have an atherosclerotic mechanism in BAD, requiring potent antiplatelet therapy for the secondary prevention of ischemic stroke.

CADMUS: A Novel MRI-Based Classification of Spontaneous Intracerebral Hemorrhage Associated With Cerebral Small Vessel Disease

Cerebral small vessel disease (SVD) is the major cause of intracerebral hemorrhage (ICH). There is no comprehensive, easily applicable classification of ICH subtypes according to the presumed underlying SVD using MRI. We developed an MRI-based classification for SVD-related ICH. We performed a retrospective study in the prospectively collected Swiss Stroke Registry (SSR, 2013-2019) and the Stroke InvestiGation in North And central London (SIGNAL) cohort. Patients with nontraumatic, SVD-related ICH and available MRI within 3 months were classified as Cerebral Amyloid angiopathy (CAA), Deep perforator arteriopathy (DPA), Mixed CAA-DPA, or Undetermined SVD using hemorrhagic and nonhemorrhagic MRI markers (CADMUS classification). The primary outcome was inter-rater reliability using Gwet's AC1. Secondary outcomes were recurrent ICH/ischemic stroke at 3 months according to the CADMUS phenotype. We performed Firth penalized logistic regressions and competing risk analyses. The SSR cohort included 1,180 patients (median age [interquartile range] 73 [62-80] years, baseline NIH Stroke Scale 6 [2-12], 45.6% lobar hematoma, systolic blood pressure on admission 166 [145-185] mm Hg). The CADMUS phenotypes were as follows: mixed CAA-DPA (n = 751 patients, 63.6%), undetermined SVD (n = 203, 17.2%), CAA (n = 154, 13.1%), and DPA (n = 72, 6.3%), with a similar distribution in the SIGNAL cohort (n = 313). Inter-rater reliability was good (Gwet's AC1 for SSR/SIGNAL 0.69/0.74). During follow-up, 56 patients had 57 events (28 ICH, 29 ischemic strokes). Three-month event rates were comparable between the CADMUS phenotypes. CADMUS, a novel MRI-based classification for SVD-associated ICH, is feasible and reproducible and may improve the classification of ICH subtypes in clinical practice and research.

MRI-Based Classification for Tibial Spine Fracture: Detection Efficacy, Classification Accuracy, and Reliability

Recently, a new MRI-based classification for evaluating tibial spine fractures (TSFs) was developed to aid in treating these injuries. Our objective was to assess the detection efficacy, classification accuracy, and reliability of this classification in detecting and grading TSFs, as well as its impact on treatment strategy, compared to the Meyers and McKeever (MM) classification. A retrospective study included 68 patients with arthroscopically confirmed TSFs. All patients had plain radiography and conventional MRI of the affected knee before arthroscopy. Three experienced radiologists independently reviewed all plain radiographs and MRI data and graded each patient according to MM and MRI-based classifications. The detection efficacy, classification accuracy, and inter-rater agreement of both classifications were evaluated and compared, using arthroscopic findings as the gold standard. The final analysis included 68 affected knees. Compared to the MM classification, the MRI-based classification produced 22.0% upgrade of TSFs and 11.8% downgrade of TSFs. According to the reviewers, the fracture classification accuracy of the MRI-based classification (91.2-95.6%) was significantly higher than that of the MM classification (73.5-76.5%, p=0.002-0.01). The fracture detection rate of MRI-based classification (94.1-98.5%) was non-significantly higher than that of the MM classification (83.8-89.7%, p=0.07-0.4). The soft tissue injury detection accuracy for MRI-based classification was 91.2-94.1%. The inter-rater reliability for grading TSFs was substantial for both the MM classification (κ=0.69) and MRI-based classification (κ=0.79). MRI-based classification demonstrates greater accuracy and reliability compared to MM classification for detecting and grading TSFs and associated soft tissue injuries.

Magnetic Resonance Imaging-Based Classification Systems for Informing Better Outcomes of Adenomyosis After Ultrasound-Guided High-Intensity Focused Ultrasound Ablating Surgery.

A referenced MRI-based classification associated with focused ultrasound ablation surgery (FUAS) outcomes is lacking in adenomyosis. To identify an MRI-based classification system for informing the FUAS outcomes. Retrospective. Patients with FUAS for adenomyosis, were divided into a training set (N = 643; 355 with post-FUAS gonadotropin-releasing hormone/levonorgestrel, 288 without post-FUAS therapy) and an external validation set (N = 135; all without post-FUAS therapy). 1.5 T, turbo spin-echo T2-weighted imaging and single-shot echo-planar diffusion-weighted imaging sequences. Five MRI-based adenomyosis classifications: classification 1 (C1) (diffuse, focal, and mild), C2 (intrinsic, extrinsic, intramural, and indeterminate), C3 (internal, adenomyomas, and external), C4 (six subtypes on areas [internal or external] and volumes [<1/3 or ≥2/3]), and C5 (internal [asymmetric or symmetric], external, intramural, full thickness [asymmetric or symmetric]) for FUAS outcomes (symptom relief and recurrence). The optimal classification was significantly associated with the most subtypes of FUAS outcomes. Relating to the timing of recurrence was measured using Cox regression analysis and median recurrence time was estimated by a Kaplan-Meier curve. A P value <0.05 was considered statistically significant. Dysmenorrhea relief and recurrence were only associated with C2 in training patients undergoing FUAS alone. Compared with other subtypes, the extrinsic subtype of C2 was significantly associated with dysmenorrhea recurrence in the FUAS group. Besides, the median dysmenorrhea recurrence time of extrinsic subtype was significantly shorter than that of other subtypes (42.0 months vs. 50.3 months). In the validation cohort, C2 was confirmed as the optimal system and its extrinsic subtype was confirmed to have a significantly shorter dysmenorrhea recurrence time than other subtypes. Classification 2 can inform dysmenorrhea relief and recurrence in patients with adenomyosis undergoing FAUS only. Itsextrinsic subtype was associated with an earlier onset of dysmenorrhea recurrence after treatment. 3 TECHNICAL EFFICACY: Stage 5.

Evaluation of a New MRI-Based Classification of Graft Status After Superior Capsule Reconstruction.

A classification system for the graft state after superior capsule reconstruction (SCR) using magnetic resonance imaging (MRI) has not been described previously. To introduce a new, MRI-based classification system for graft integrity after SCR and to evaluate the system according to postoperative outcomes. Cohort study (diagnosis); Level of evidence, 3. Included were 62 consecutive patients who underwent SCR using autologous fascia lata graft between January 2013 and April 2021. Postoperative outcomes were assessed (American Shoulder and Elbow Surgeons [ASES] score, Constant score, pain visual analog scale [pVAS], range of motion [ROM], acromiohumeral distance [AHD], Hamada grade). Graft status was classified by 2 orthopaedic surgeons on postoperative MRI in accordance with the signal intensity and the presence or extent of the tear, as follows: type 1 (hypointense signal without tear), type 2 (hyperintense signal without tear), type 3 (partial-thickness tear), type 4 (full-thickness tear with partial continuity), and type 5 (full-thickness tear with complete discontinuity). Intra- and interobserver agreement were assessed using Cohen kappa. The correlation between postoperative outcomes (ASES score, Constant score, pVAS, ROM, AHD, and Hamada grade) and the SCR graft classification system was assessed with the Pearson correlation coefficient, and the outcomes were compared according to classification type. Patients were classified according to the new system as follows: type 1 (n = 15), type 2 (n = 20), type 3 (n = 7), type 4 (n = 8), and type 5 (n = 12). There was excellent interobserver agreement (κ = 0.819) and intraobserver agreement (κ = 0.937 and 0.919). The classification system showed a moderate to high correlation with the ASES score (r = -0.451; P = .001), pVAS (r = 0.359; P = .005), AHD (r = -0.642; P < .001), and Hamada grade (r = 0.414; P < .001). Patients classified as having types 1 and 2 showed better outcomes in terms of ASES score, pVAS, ROM, and AHD compared with type 5 patients (P ≤ .021 for all). The new classification system was highly reproducible and showed clinical utility for both radiological and clinical evaluation after SCR.

A robust MRI-based brain tumor classification via a hybrid deep learning technique

The brain is the most vital component of the neurological system. Therefore, brain tumor classification is a very challenging task in the field of medical image analysis. There has been a qualitative leap in the field of artificial intelligence, deep learning, and their medical imaging applications in the last decade. The importance of this remarkable development has emerged in the field of biomedical engineering due to the sensitivity and seriousness of the issues related to it. The use of deep learning in the field of detecting and classifying tumors in general and brain tumors in particular using magnetic resonance imaging (MRI) is a crucial factor in the accuracy and speed of diagnosis. This is due to its great ability to deal with huge amounts of data and avoid errors resulting from human intervention. The aim of this research is to develop an efficient automated approach for classifying brain tumors to assist radiologists instead of consuming time looking at several images for a precise diagnosis. The proposed approach is based on 3064 T1-weighted contrast-enhanced brain MR images (T1W-CE MRI) from 233 patients. In this study, the proposed system is based on the results of five different models to use the combined potential of multiple models, trying to achieve promising results. The proposed system has led to a significant improvement in the results, with an overall accuracy of 99.31%.

A New Breakpoint to Classify 3D Voxels in MRI: A Space Transform Strategy with 3t2FTS-v2 and Its Application for ResNet50-Based Categorization of Brain Tumors.

Three-dimensional (3D) image analyses are frequently applied to perform classification tasks. Herein, 3D-based machine learning systems are generally used/generated by examining two designs: a 3D-based deep learning model or a 3D-based task-specific framework. However, except for a new approach named 3t2FTS, a promising feature transform operating from 3D to two-dimensional (2D) space has not been efficiently investigated for classification applications in 3D magnetic resonance imaging (3D MRI). In other words, a state-of-the-art feature transform strategy is not available that achieves high accuracy and provides the adaptation of 2D-based deep learning models for 3D MRI-based classification. With this aim, this paper presents a new version of the 3t2FTS approach (3t2FTS-v2) to apply a transfer learning model for tumor categorization of 3D MRI data. For performance evaluation, the BraTS 2017/2018 dataset is handled that involves high-grade glioma (HGG) and low-grade glioma (LGG) samples in four different sequences/phases. 3t2FTS-v2 is proposed to effectively transform the features from 3D to 2D space by using two textural features: first-order statistics (FOS) and gray level run length matrix (GLRLM). In 3t2FTS-v2, normalization analyses are assessed to be different from 3t2FTS to accurately transform the space information apart from the usage of GLRLM features. The ResNet50 architecture is preferred to fulfill the HGG/LGG classification due to its remarkable performance in tumor grading. As a result, for the classification of 3D data, the proposed model achieves a 99.64% accuracy by guiding the literature about the importance of 3t2FTS-v2 that can be utilized not only for tumor grading but also for whole brain tissue-based disease classification.

Three-Grade Magnetic Resonance Imaging–Based Gluteus Medius and/or Minimus Tear Classification System Provides Excellent Inter-Rater Reliability

To develop a magnetic resonance imaging (MRI)-based classification system integrating tear characteristics including tear thickness (partial vs full) and tear retraction (less than or greater than 2 cm) for gluteus medius and/or minimus tears and to determine the inter-rater reliability of this MRI-based classification for gluteus medius and/or minimus tears. Patients who underwent primary endoscopic or open repair of gluteus medius and/or minimus tears between 2012 and 2022 were identified to be included in the review of 1.5-T MRI scans. One hundred MRI scans were randomized for review by 2 orthopaedic surgeons and evaluated for tear thickness (partial vs full), extent of retraction, and degree of fatty infiltration according to an applied Goutallier-Fuchs (G-F) classification. Tears were also graded according to the 3-grade MRI-based classification system as follows: grade 1, partial-thickness tears; grade 2, full-thickness tears with less than 2 cm of retraction; grade 3, full thickness with 2 cm or more retraction. Inter-rater reliability was calculated by absolute and relative agreement using Cohen's kappa (κ). Significance was defined by P value <.05. In total, 221 patients were identified, and after application of exclusion criteria and randomization, 100 scans were evaluated. The 3-grade classification system demonstrated high absolute agreement (88%) comparable to the absolute agreement of the G-F classification (67%). The 3-grade classification system demonstrated substantial inter-rater reliability (κ= 0.753), whereas the G-F classification demonstrated moderate inter-rater reliability (κ= 0.489). The proposed 3-grade MRI-based classification system for gluteus medius and/or minimus tears demonstrated substantial inter-rater reliability, comparable with that of the applied G-F classification. It is important to understand how gluteus medius and/or minimus tear characteristics impact postoperative outcomes. The 3-grade MRI-based classification incorporates tear thickness and amount of retraction that can complement previous classification systems to give the provider and patient more information when considering treatment options.

Anatomical variations of the cavernous Internal Carotid Artery: Validation of an MRI-based classification

ObjectiveVariations of the cavernous Internal Carotid Artery (ICA) angulation (C4-bend) have been classified into 4 anatomical subtypes with particular surgical relevance, as a very angulated ICA comes in closest contact with the pituitary gland leading to higher risk of iatrogenic vascular injury. This study aimed to validate this classification using current routine imaging techniques. MethodsThe different cavernous ICA bending angles were measured on 109 MRI TOF sequences, within a retrospective database of patients with no sellar lesions. Each ICA was classified into one of 4 anatomical subtypes as already defined in a previous study [1]. Interrater agreement was assessed by a Kappa Correlation Coefficient. ResultsKappa Correlation Coefficient was of 0.90 [0.82–0.95], validating the strong level of agreement between all observers when using the present classification. ConclusionClassification of the cavernous ICA in 4 subtypes appears statistically valid on routinely used pre-operative MRI, providing an efficient tool to estimate the vascular iatrogenic risk prior to endoscopic endonasal transsphenoidal surgery.

Wrist &amp; Hand

The April 2023 Wrist &amp; Hand Roundup360 looks at: MRI-based classification for acute scaphoid injuries: the OxSMART; Deep learning for detection of scaphoid fractures?; Ulnar shortening osteotomy in adolescents; Cost-utility analysis of thumb carpometacarpal resection arthroplasty; Arthritis of the wrist following scaphoid fracture nonunion; Extensor hood injuries in elite boxers; Risk factors for reoperation after flexor tendon repair; Nonoperative versus operative treatment for displaced finger metacarpal shaft fractures.

MRI-based Brain Tumor Image Classification Using CNN

Though all brain tumors are not cancerous but they caused a critical disease produced by irrepressible and unusual dividing of cells. For the case of Medical diagnostics of many diseases, the health industry needs help, the current development in the arena of deep learning has assisted to detect diseases. In recent years medical image classification has gained remarkable attention. The most well-known neural network model for image classification problems is the Convolutional Neural Network (CNN). CNN is the frequently employed machine-learning algorithm that is used in Visual learning and Image Recognition research. It is considered to derive features adaptively through convolution, activation, pooling, and fully connected layers. In our paper, we present the convolutional neural network method to determine cancerous and non-cancerous brain tumors. We also used Data Augmentation and Image Processing to classify brain (Magnetic Resonance Imaging (MRI). We used two significant steps in our proposed system. First, different image processing techniques are used to preprocess the images and secondly we classify the preprocessed image using CNN. Brain tumor classification is a process of identifying and separating the cancerous and non-cancerous brain tissues and labeling them automatically. We use the famous machine learning algorithms Convolutional Neural Network which is broadly employed for image classifications. This experiment is conducted on a dataset of 2065 images. In our dataset number of training, examples are 1445, the number of validation examples is 310, and the number of testing example is 310. We also used data augmentation to raise the number of the dataset. We achieved a high testing accuracy of 94.39%. The proposed system displayed sufficient accuracy on the dataset and beat many of the noticeable present methods.

Optimizing MRI-based brain tumor classification and detection using AI: A comparative analysis of neural networks, transfer learning, data augmentation, and the cross-transformer network

Early detection and diagnosis of brain tumors are crucial to taking adequate preventive measures, as with most cancers. On the other hand, artificial intelligence (AI) has grown exponentially, even in such complex environments as medicine. Here it’s proposed a framework to explore state-of-the-art deep learning architectures for brain tumor classification and detection. An own development called Cross-Transformer is also included, which consists of three scalar products that combine self-care model keys, queries, and values. Initially, we focused on the classification of three types of tumors: glioma, meningioma, and pituitary. With the Figshare brain tumor dataset was trained the InceptionResNetV2, InceptionV3, DenseNet121, Xception, ResNet50V2, VGG19, and EfficientNetB7 networks. Over 97 % of classifications were accurate in this experiment, which provided a network’s performance overview. Subsequently, we focused on tumor detection using the Brain MRI Images for Brain Tumor Detection and The Cancer Genome Atlas Low-Grade Glioma database. The development encompasses learning transfer, data augmentation, as well as image acquisition sequences; T1-weighted images (T1WI), T1-weighted post-gadolinium (T1-Gd), and Fluid-Attenuated Inversion Recovery (FLAIR). Based on the results, using learning transfer and data augmentation increased accuracy by up to 6 %, with a p-value below the significance level of 0.05. As well, the FLAIR sequence was the most efficient for detection. As an alternative, our proposed model proved to be the most effective in terms of training time, using approximately half the time of the second fastest network.

MRI-based classification of IDH mutation and 1p/19q codeletion status of gliomas using a 2.5D hybrid multi-task convolutional neural network.

IDH mutation and 1p/19q codeletion status are important prognostic markers for glioma that are currently determined using invasive procedures. Our goal was to develop artificial intelligence-based methods to noninvasively determine molecular alterations from MRI. Pre-operative MRI scans of 2648 glioma patients were collected from Washington University School of Medicine (WUSM; n = 835) and publicly available Brain Tumor Segmentation (BraTS; n = 378), LGG 1p/19q (n = 159), Ivy Glioblastoma Atlas Project (Ivy GAP; n = 41), The Cancer Genome Atlas (TCGA; n = 461), and the Erasmus Glioma Database (EGD; n = 774) datasets. A 2.5D hybrid convolutional neural network was proposed to simultaneously localize glioma and classify its molecular status by leveraging MRI imaging features and prior knowledge features from clinical records and tumor location. The models were trained on 223 and 348 cases for IDH and 1p/19q tasks, respectively, and tested on one internal (TCGA) and two external (WUSM and EGD) test sets. For IDH, the best-performing model achieved areas under the receiver operating characteristic (AUROC) of 0.925, 0.874, 0.933 and areas under the precision-recall curves (AUPRC) of 0.899, 0.702, 0.853 on the internal, WUSM, and EGD test sets, respectively. For 1p/19q, the best model achieved AUROCs of 0.782, 0.754, 0.842, and AUPRCs of 0.588, 0.713, 0.782, on those three data-splits, respectively. The high accuracy of the model on unseen data showcases its generalization capabilities and suggests its potential to perform "virtual biopsy" for tailoring treatment planning and overall clinical management of gliomas.

MRI-Based Classification of Rathke’s Cleft Cyst and Its Clinical Implication

Rathke's cleft cysts (RCCs) are benign tumors of the pituitary gland. Small, asymptomatic RCCs do not require surgical treatment, whereas surgical treatment is required for symptomatic RCCs. We retrospectively reviewed medical records of patients with an RCC who were diagnosed and managed in our institution between April 2004 and April 2020 and generated two different cohorts: the observation (n=114) and the surgical group (n=99). Their initial MRI signal characteristics were analyzed. The natural course focusing on cyst size was observed in the observation group and postoperative visual and endocrine outcomes were evaluated in the surgical group. The characterization of MRI signals of cyst contents in both T1-weighted (T1W) and T2-weighted (T2W) images revealed nine combinations for our 213 patients. Among 115 patients with a high T2W signal, the cysts showed hypo-, iso-, and hyper-intensity on T1W images in 72, 39, and 44 patients, respectively; Type S-low, Type S-iso, and Type S-high. One more major group of 35 patients showed RCCs with hyperintensity on the T1W images and hypointensity on the T2W images named as Type M. In the comparison between observation and surgical groups, we identified only two major groups in which the number of patients in the surgical and observation groups was statistically different: more Type S-low in a surgical group (p<0.001) and more Type M in an observation group (p=0.007). In subgroup analysis, the range of change in the cyst size was the highest in Type S-high in the observation group (p=0.028), and intergroup differences in visual and endocrine outcomes were not evident in the surgical group. MRI characteristics help to predict the natural course of RCCs. We identified subgroups of RCCs which are more or less likely to require surgical intervention.

A deep learning framework integrating MRI image preprocessing methods for brain tumor segmentation and classification.

Glioma grading is critical in treatment planning and prognosis. This study aims to address this issue through MRI-based classification to develop an accurate model for glioma diagnosis. Here, we employed a deep learning pipeline with three essential steps: (1) MRI images were segmented using preprocessing approaches and UNet architecture, (2) brain tumor regions were extracted using segmentation, then (3) high-grade gliomas and low-grade gliomas were classified using the VGG and GoogleNet implementations. Among the additional preprocessing techniques used in conjunction with the segmentation task, the combination of data augmentation and Window Setting Optimization was found to be the most effective tool, resulting in the Dice coefficient of 0.82, 0.91, and 0.72 for enhancing tumor, whole tumor, and tumor core, respectively. While most of the proposed models achieve comparable accuracies of about 93% on the testing dataset, the pipeline of VGG combined with UNet segmentation obtains the highest accuracy of 97.44%. In conclusion, the presented architecture illustrates a realistic model for detecting gliomas; moreover, it emphasizes the significance of data augmentation and segmentation in improving model performance.

Brain tumor classification based on neural architecture search

Brain tumor is a life-threatening disease and causes about 0.25 million deaths worldwide in 2020. Magnetic Resonance Imaging (MRI) is frequently used for diagnosing brain tumors. In medically underdeveloped regions, physicians who can accurately diagnose and assess the severity of brain tumors from MRI are highly lacking. Deep learning methods have been developed to assist physicians in detecting brain tumors from MRI and determining their subtypes. In existing methods, neural architectures are manually designed by human experts, which is time-consuming and labor-intensive. To address this problem, we propose to automatically search for high-performance neural architectures for classifying brain tumors from MRIs, by leveraging a Learning-by-Self-Explanation (LeaSE) architecture search method. LeaSE consists of an explainer model and an audience model. The explainer aims at searching for a highly performant architecture by encouraging the architecture to generate high-fidelity explanations of prediction outcomes, where explanations’ fidelity is evaluated by the audience model. LeaSE is formulated as a four-level optimization problem involving a sequence of four learning stages which are conducted end-to-end. We apply LeaSE for MRI-based brain tumor classification, including four classes: glioma, meningioma, pituitary tumor, and healthy, on a dataset containing 3264 MRI images. Results show that our method can search for neural architectures that achieve better classification accuracy than manually designed deep neural networks while having fewer model parameters. For example, our method achieves a test accuracy of 90.6% and an AUC of 95.6% with 3.75M parameters while the accuracy and AUC of a human-designed network—ResNet101—is 84.5% and 90.1% respectively with 42.56M parameters. In addition, our method outperforms state-of-the-art neural architecture search methods.

Performance enhancement of MRI-based brain tumor classification using suitable segmentation method and deep learning-based ensemble algorithm

Glioma is the most common brain tumor in humans. Accurate stage estimation of the tumor is essential for treatment planning. The biopsy is the gold standard method for this purpose. However, it is an invasive procedure, which can prove fatal for patients, if a tumor is present deep inside the brain. Therefore, a magnetic resonance imaging (MRI) based non-invasive method is proposed in this paper for low-grade glioma (LGG) versus high-grade glioma (HGG) classification. To maximize the above classification performance, five pre-trained convolutional neural networks (CNNs) such as AlexNet, VGG16, ResNet18, GoogleNet, and ResNet50 are assembled using a majority voting mechanism. Segmentation methods require human intervention and additional computational efforts. It makes computer-aided diagnosis tools semi-automated. To analyze the performance effect of segmentation methods, three segmentation methods such as region of interest MRI segmentation (RSM) and skull-stripped MRI segmentation (SSM), and whole-brain MRI (WBM) (non-segmentation) data were compared using above mentioned algorithm. The highest classification accuracy of 99.06 ± 0.55 % was observed on the RSM data and the lowest accuracy of 98.43 ± 0.89 % was observed on the WSM data. However, only a 0.63 % improvement was found in the accuracy of the RSM data against the WBM data. This shows that deep learning models have an incredible ability to extract appropriate features from images. Furthermore, the proposed algorithm showed 2.85 %, 1.39 %, 1.26 %, 2.66 %, and 2.33 % improvement in the average accuracy of the above three datasets over the AlexNet, VGG16, ResNet18, GoogleNet, and ResNet50 models, respectively.

Structural Correlates of Overt Sentence Reading in Mild Cognitive Impairment and Mild-to-Moderate Alzheimer's Disease.

Overt sentence reading in mild cognitive impairment (MCI) and mild-tomoderate Alzheimer's disease (AD) has been associated with slowness of speech, characterized by a higher number of pauses, shorter speech units and slower speech rate and attributed to reduced working memory/ attention and language capacity. This preliminary case-control study investigates whether the temporal organization of speech is associated with the volume of brain regions involved in overt sentence reading and explores the discriminative ability of temporal speech parameters and standard volumetric MRI measures for the classification of MCI and AD. Individuals with MCI, mild-to-moderate AD, and healthy controls (HC) had a structural MRI scan and read aloud sentences varying in cognitive-linguistic demand (length). The association between speech features and regional brain volumes was examined by linear mixed-effect modeling. Genetic programming was used to explore the discriminative ability of temporal and MRI features. Longer sentences, slower speech rate, and a higher number of pauses and shorter interpausal units were associated with reduced volumes of the reading network. Speech-based classifiers performed similarly to the MRI-based classifiers for MCI-HC (67% vs. 68%) and slightly better for AD-HC (80% vs. 64%) and AD-MCI (82% vs. 59%). Adding the speech features to the MRI features slightly improved the performance of MRI-based classification for AD-HC and MCI-HC but not HC-MCI. The temporal organization of speech in overt sentence reading reflects underlying volume reductions. It may represent a sensitive marker for early assessment of structural changes and cognitive- linguistic deficits associated with healthy aging, MCI, and AD.

Decomposition-Based Correlation Learning for Multi-Modal MRI-Based Classification of Neuropsychiatric Disorders.

Multi-modal magnetic resonance imaging (MRI) is widely used for diagnosing brain disease in clinical practice. However, the high-dimensionality of MRI images is challenging when training a convolution neural network. In addition, utilizing multiple MRI modalities jointly is even more challenging. We developed a method using decomposition-based correlation learning (DCL). To overcome the above challenges, we used a strategy to capture the complex relationship between structural MRI and functional MRI data. Under the guidance of matrix decomposition, DCL takes into account the spike magnitude of leading eigenvalues, the number of samples, and the dimensionality of the matrix. A canonical correlation analysis (CCA) was used to analyze the correlation and construct matrices. We evaluated DCL in the classification of multiple neuropsychiatric disorders listed in the Consortium for Neuropsychiatric Phenomics (CNP) dataset. In experiments, our method had a higher accuracy than several existing methods. Moreover, we found interesting feature connections from brain matrices based on DCL that can differentiate disease and normal cases and different subtypes of the disease. Furthermore, we extended experiments on a large sample size dataset and a small sample size dataset, compared with several other well-established methods that were designed for the multi neuropsychiatric disorder classification; our proposed method achieved state-of-the-art performance on all three datasets.