Carotid B-mode ultrasound (U/S) video analysis enables the detection of temporal changes of plaque features, which may help identify patients at increased risk of ischemic stroke. Despite existing evidence, whole plaque and intraplaque area motion has not been widely investigated in 2D B-mode U/S videos. Here, we collected, filtered and discussed previous studies proposing different methodologies for carotid plaque motion analysis in 2D B-mode U/S videos. Our main objective was to detect whether robust and reproducible methods exist for plaque motion analysis in 2D longitudinal B-mode U/S videos, obtained in a clinical setting. Also, whether any of them has been considered to assist doctors in detecting high-risk plaques, based on the analysis of all intraplaque areas. We searched for eminent methodological gaps, possibly evoking the necessity for further analysis of the individual plaque components. We searched on PubMed, Scopus, and IEEE, from 2012 to 2024, considering studies that fulfilled the following criteria: a. inclusion of 2D carotid U/S B-mode longitudinal image sequences or videos, b. inclusion of Asymptomatic (AS) and Symptomatic (SY) patients with carotid stenosis ≥50%, and c. studies involving developed carotid plaques (not intima-media thickness). We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR 2020) guidelines. Notably, unanswered questions still remain, such as how motion should be evaluated in cases with >1 plaque area and how the (set of) specific intraplaque areas, whose discordant motion might be heavily involved in plaque rupture in AS or SY cases, could be identified.

This study aimed to analyze 2D B-mode longitudinal carotid ultrasound videos to identify and characterize the motion of plaque compositions. Understanding these motion patterns in both asymptomatic (AS) and symptomatic (SY) cases may provide valuable insights into plaque instability and its potential role in rupture and ischemic stroke. We included 20 carotid ultrasound videos (10 AS and 10 SY patient videos). First, a previously trained and evaluated Deep Learning model segmented the whole plaque region of interest (ROIs), per video frame. Based on a heuristic computational approach, we contoured each ROI into 6 colors (6 grayscale, GS intensity ranges), depicting different histological compositions (e.g., calcified areas, areas with fibrous content and/or collagen, and lipid cores). We estimated motion of the whole plaque and that of each composition, across 3 systoles per video, calculating the Maximum Angular Spread (MAXFW) and the maximum median motion magnitude. The GS≤25 areas in highly-discordant AS and SY plaques exhibited the largest MAXFW (moderate-discordant motion), compared to other compositions. The MAXFW of the GS≤25 in concordant AS plaques, although higher compared to others, showed a concordant motion. Overall, the angular and the motion spread both showed a continuous increase from the brighter to the darkest areas in the highly-discordant AS and SY plaques. These findings might reflect a key involvement of the echolucent plaque areas in the development of discordant motion in both the AS and SY highly-discordant plaques. This could possibly help doctors detect high-risk areas more prone to rupture, in 2D B-mode longitudinal carotid ultrasound videos.Clinical Relevance— This study highlights the importance of estimating the motion of each composition present in the carotid plaques, when examining 2D B-mode ultrasound videos, to more effectively identify plaque areas more likely to rupture.

In this study we propose and evaluate AtheroRisk, a standalone integrated computer software system for the analysis of carotid B-mode ultrasound (U/S) images and videos. Our goal was to provide a tool to help physicians stratify stroke risk. The presented system is based on outcomes from different research studies, as well as European guidelines for the evaluation and treatment of carotid artery disease. The goal is to enable reproducible analysis of atherosclerotic plaques in real time. AtheroRisk brings together analysis of U/S image and/or video of the carotid arteries facilitating anonymization, standardization, noise filtering and segmentation of plaques; followed by image and motion feature extraction to derive plaque-composition and stability. All the above steps are combined to determine the annual stroke risk rate and the 5-year stroke-free survival rate. The system is powered by an SQLite local database, in which the end user can save and manage data extracted from processing and analysis in an anonymized way. The first version of AtheroRisk was developed following an incremental software development process. The developed system was verified using 54 U/S videos (27 Asymptomatic, AS; 27 Symptomatic, SY cases) while the clinician’s satisfaction and comments were collected through a questionnaire-based validation process. The results show that the integrated system proposed in this study can be successfully used for the automated image and video analysis of the CCA plaque in ultrasound videos.

Multiple Sclerosis (MS) is a chronic autoimmune disease that affects the central nervous system. Nerves are covered by a layer called myelin, which is responsible for protecting and maintaining their functionality. In the case of MS, the myelin becomes damaged, resulting in the nerves functioning unpredictably. MS is a disease characterized by relapses, which can cause permanent and irreversible damage to the patient's mobility, vision, and sensations, unless appropriate medical treatment is administered in time. This highlights the importance of the early detection of the presence or differentiation of MS using computer assisted diagnosis (CAD) systems. This study proposes an automatic segmentation approach of MS damaged regions using deep neural networks that can be integrated with any 3rd party CAD system. The data utilized comprised of a total of 1838 T2-type Magnetic Resonance Imaging (MRI) images collected from a cohort of 38 patients who underwent imaging at two distinct time points. Each MRI image was accompanied by meta-data, detailing the specific locations of the observed damage attributed to MS. The experimental setup investigated and optimized a comprehensive combination of different data preprocessing and augmentation steps, using several variations of the U-Net convolutional neural network (CNN), like U-Net++, Attention U-Net, ResUNet-a and TransUNet. The proposed methods achieved a segmentation accuracy of 0.70 based on the Dice Similarity Coefficient (DSC), a performance comparable to 2D and 3D state-of-the-art approaches in literature. A software framework encapsulating the automated model was also developed to facilitate the clinical practice workflow and underpin adoption by medical practitioners.

Ultrasound analysis of the diaphragm can provide valuable insights into diaphragmatic function in neonates, potentially enabling physicians to identify disease-related abnormalities. We propose and evaluate an integrated semi-automated video analysis system for the accurate diaphragmatic motion analysis in neonates. In order to enhance the clinical learning outcome and precisely assess the suggested system, we employed 20 simulated ultrasound videos of the diaphragm that were produced using characteristics of typical neonate’s diaphragmatic motion. To evaluate the proposed system, manual (performed by a doctor) vs automated measurements were extracted and diaphragmatic excursion (DE), inspiration time (Tinsp), total breathing time (Ttot), diaphragmatic curve slope (DSP), and relaxation rate (RR) were measured. The following manual (–/) vs semi-automated (/–), (median ± IQR) measurements were computed for all simulated videos; The correlation coefficient (ρ, p-value) are also given showing the correlation of the parameters measured by the doctor vs the proposed system: (i) DE: (3.66 ± 0.11)/3.71 ± 0.24 mm (ρ = 0.97, p = 0.002), (ii) Tinsp: (0.55 ± 0.19)/(0.57 ± 0.16) sec (ρ = 0.91, p = 0.002), (iii) Ttot: (1.31 ± 0.19)/(1.31 ± 0.18) sec (ρ = 0.90, p = 0.005), (iv) DSC: 6.69 ± 2.73/6.58 ± 2.31 mm/sec (ρ = 0.79, p = 0.003), (v) RR: 4.88 ± 0.15/5.08 ± 0.57 mm/sec (ρ = 0.89, p = 0.001). This is the first study reported in the literature that utilizes a semi-automated integrated motion analysis system to assess diaphragmatic motion parameters in neonate simulation videos. No statistically significant differences were found between manual and semi-automated measurements. Therefore, the proposed technique could be helpful for the clinical evaluation of neonate’s diaphragmatic motion. Real ultrasound videos of neonate diaphragms in both normal and pathological motion will be used for additional research and validation on a larger sample.

This study proposes an automated segmentation of prostate cancer in transrectal ultrasound images using different preprocessing methods to enhance the segmentation accuracy. We propose the use of image intensity normalization and despeckle filtering, individually and in combination, as preprocessing techniques to improve the performance of a deep learning segmentation model (DeepLabv3 +) in ultrasound images of prostate cancer. This algorithm was applied to a dataset of 647 TRUS images. All images were separated into four groups as follows: original (O), intensity normalized (N), despeckled (D), and intensity normalized and despeckled (ND). Manual segmentations of the prostate were performed by an experienced radiation oncologist and compared with automated segmentations using six different evaluation metrics. Statistical analysis showed that preprocessing enhances segmentation performance, with a median (±IQR) Dice coefficient of 94.02 (3.93)/94.84 (3.92)/94.43 (3.05)/94.22 (4.19) for the O/N/D/ND images respectively. The highest segmentation accuracy was achieved on the N images, followed by the ND images which confirm the benefits of N and ND in enhancing the final segmentation accuracy. No statistically significant differences were found between all different preprocessing schemes for all the evaluation metrics investigated. Due to the small number of patients, the generalizability of the results is limited. Nevertheless, the findings highlight the potential clinical value of preprocessing in improving segmentation performance in challenging ultrasound cases. Additional experimentation with a larger image dataset and other alternative evaluation metrics is required to validate the present results.

Prostate cancer (PCa) is a major global health concern for men and the ability to detect it in its early stages is important. While imaging modalities such as Transrectal Ultrasound (TRUS) constitutes a critical role in diagnosis, challenges such as noise and limited specificity hinder their effectiveness especially when features are extracted from the images which may be used for classification of cancer. This study investigates the impact of various preprocessing techniques, including ultrasound image normalization (N), despeckle filtering (D), and normalization and despeckle filtering (ND) on texture features. We seek to improve the diagnostic precision of PCa by using the variability in texture features taken from the prostate. Image normalization and despeckling methods were employed, where image quality was evaluated using four different evaluation metrics (EM) and a large number of texture features extracted from the automated segmented prostate area. Statistical analyses were used to assess the stability and diagnostic reliability of texture features extracted under different preprocessing schemes. A number of features demonstrated robustness, whereas others exhibited larger variability. This study confirmed the advantages of N, D and ND in improving the image quality and stability of features in PCa ultrasound images. Additional experimentation with a larger image dataset and other alternative evaluation metrics is required to validate the present results.

Carotid B-mode ultrasound (U/S) imaging provides more than the degree of stenosis in stroke risk assessment. Plaque morphology and texture have been extensively investigated in U/S images, revealing plaque components, such as juxtaluminal black areas close to lumen (JBAs), whose size is linearly related to the risk of stroke. Convolutional neural networks (CNNs) have joined the battle for the identification of high-risk plaques, although the ways they perceive asymptomatic (ASY) and symptomatic (SY) plaque features need further investigation. In this study, the objective was to assess whether class activations maps (CAMs) can reveal which U/S grayscale-(GS)-based plaque compositions (lipid cores, fibrous content, collagen, and/or calcified areas) influence the model's understanding of the ASY and SY cases. We used Xception via transfer learning, as a base for feature extraction (all layers frozen), whose output we fed into a new dense layer, followed by a new classification layer, which we trained with standardized B-mode U/S longitudinal plaque images. From a total of 236 images (118 ASY and 118 SY), we used 168 in training (84 ASY and 84 SY), 22 in internal validation (11 ASY and 11 SY), and 46 in testing (23 ASY and 23 SY). In testing, the model reached an accuracy, sensitivity, specificity, and area under the curve at 80.4%, 82.6%, 78.3%, and 0.80, respectively. Precision and the F1 score were found at 81.8% and 80.0%, and 79.2% and 80.9%, for the ASY and SY cases, respectively. We used faster-Score-CAM to produce a heatmap for each tested image, quantifying each plaque composition area overlapping with the heatmap to find compositions areas related to ASY and SY cases. Dark areas (GS ≤ 25) or JBAs (whose presence was verified priorly, by an experienced vascular surgeon) were found influential for the understanding of both the ASY and the SY plaques. Calcified areas, fibrous content, and lipid cores, together, were more related to ASY plaques. These findings indicate the need for further investigation on how the GS ≤ 25 plaque areas affect the learning process of the CNN models, and they will be further validated.

The objective of this work was to investigate a new sparse multiscale Amplitude Modulation - Frequency Modulation (AM-FM) analysis based on multiple Gabor filterbanks representations where component selection was carried out using the elastic net regularization equation. The AM-FM histogram features sets of instantaneous amplitude, instantaneous phase and the magnitude of instantaneous frequency were computed from carotid plaque ultrasound images to assess the risk of stroke. A total of 100 carotid plaque ultrasound images (50 asymptomatic and 50 symptomatic) were analyzed following manual segmentation by an expert. Classification modelling was carried out using the Support Vectors Machine to classify asymptomatic versus symptomatic plaques. An overall classification accuracy of 74% was achieved, demonstrating that the new sparse multiscale AM-FM analysis provided robust features. These findings are comparable with classification models trained with traditional AM-FM feature sets as well as classical texture feature sets. Moreover, the proposed analysis provides new sparse image representations that allow us to reduce the number of AM-FM components needed to explain the local spatial-frequency content and can further facilitate the desired explanatory interpretation in stroke risk assessment.

The objective of this work was to investigate the Amplitude Modulation - Frequency Modulation (AM- FM) texture feature variability in carotid ultrasound video during the cardiac cycle at systole and diastole. The goal here was to identify AM-FM features that are associated with increased risk of stroke. We computed the instantaneous amplitude, instantaneous phase and the magnitude of instantaneous frequency to extract plaque histogram features. A small dataset of 5 asymptomatic and 3 symptomatic videos were analyzed. Selected AM-FM plaque histogram texture features extracted during the cardiac cycle at the systolic and diastolic states were statistically significantly different between asymptomatic and symptomatic videos. However, further evaluation with more subjects needs to be carried out to exploit the usefulness of the proposed analysis in the clinical context.

The potential for stroke risk in humans due to clinical cardiovascular disease (CVD), which leads to the hardening of the artery walls (atherosclerosis), can be assessed by examining the common carotid artery (CCA) in ultrasound images. Specifically, this can be done by measuring the intima media thickness (IMT), which represents the thickness of the arteries wall, and by analyzing texture features (TFs) of the CCA's intima-media complex (IMC) of the artery wall. In this paper, a sample of 612 longitudinal-section ultrasound images of the left and the right CCA from 158 men and 148 women, out of which 42 demonstrated clinical CVD, is studied. All images are intensity normalized and despeckled, with the IMC segmented through an in-house system of semi-automated segmentation, where the IMT is measured, and 40 different TFs are extracted. Then, employing structural equation modeling (SEM), these TFs, which are put in 6 groups (constructs), are collectively tested on how they are related to CVD. The influence of the IMT is also studied through a moderation analysis, while gender and age of the sub-jects of the study are tested with regard to their control effect. The main conclusions of the study are as follows. The 6 TFs groups and the IMT fit the measurement model very well. Also, 6 hypothesized paths for the impact of each TFs group on CVD are tested in a structural model, with 5 of them, namely Statistical Features (SF), Spatial Gray Level Dependence Matrices (SGLDM), Gray Level Difference Statistics (GLDS), Statistical Feature Matrix (SFM) and Laws Texture Energy Measures (LTEM), impacting CVD in a moderate manner (p-values between 0.04 and 0.07). The IMT is shown to play a strong moderating role (p-values of Delta chi(2) < 0.01), enhancing the level of impact of the TFs on CVD. Gender and age have a strong controlling effect on CVD (p-values of < 0.01). The precision of the findings of this study is considerably improved compared to those previously reported, because of the very good model fit (e.g., normed fit index (NFI) = 0.94) for both measurement and structural models. However, due to the moderate nature of the impact, supplementary work is required for testing either additional combinations of TFs, as well as testing other samples through the proposed model, for further evaluation of the CVD risk on symptomatic subjects at risk of atherosclerosis.

BACKGROUND AND OBJECTIVE:Carotid B-mode ultrasound (CBUS) imaging is often used to detect and assess atherosclerotic plaques. Doctors often need to segment plaques in the CBUS images to further examine them. Multiple studies have proposed two-dimensional CBUS plaque segmentation deep learning (DL)-based solutions, achieving promising results. In most of these studies, image standardization is not reported, while not all plaque types are represented. However, prior multiple studies have highlighted the importance of data standardization in computerized CBUS plaque classification or segmentation solutions. In this study, we propose and separately evaluate three progressive preprocessing schemes, to discover the most optimal to standardize CBUS images for DL-based carotid plaque segmentation, while we also assess the effect of each preprocessing in the segmentation performance per echodensity-based plaque type (I, II, III, IV and V). METHODS:We included three CBUS image datasets (276 CBUS images, from three medical centres), with which we produced 3 data folds (with the best possible equal inclusion of images from all centers per fold), to perform 3-fold cross validation-based training and evaluation of the pre-released Channel-wise Feature Pyramid Network for Medicine (CFPNet-M) model, in carotid plaque type segmentation. We included the three data folds in their original version (O), generating also three preprocessed versions of them, namely, the resolution-normalized (R), the resolution- and intensity-normalized (RN), and the resolution- and intensity-normalized combined with despeckling (RND) versions. The samples were cropped to the plaque level, and the intersection over union (IoU) and the Dice Similarity Coefficient (DSC), along with other metrics, were used to measure the model's performance. In each training round, 12 % of the images in the 2 training folds was used for internal validation (last fold was used in evaluation). Two experienced ultrasonographers manually delineated plaques in the dataset, to provide us with ground truths, while the plaque types (I to V) were extracted according to the Gray-Weale and Geroulakos classification system. We measured the mean±standard deviation of DSC within and across the three evaluated folds, per preprocessing scheme and per plaque type. RESULTS:CFPNet-M segmented the plaques in the CBUS images in all the data preprocessing versions, yielding progressively improved performances (mean DSC at 81.9 ± 9.1 %, 83.6 ± 9.0 %, 84.1 ± 8.3 %, and 84.4 ± 8.1 % for the O, R, RN and RND 3-fold cross validation processes, respectively), irrespective of the plaque type. Interestingly, CFPNet_M yielded improved performances, for all plaque types (I, II, III, IV and V), when trained and tested with the RND data versus the O version, achieving an 80.6 ± 11 % versus 77.6 ± 17 % DSC for type I, an 84.3 ± 8 % versus 81.2 ± 9 % DSC for type II, an 84.9 ± 7 % versus 82.6 ± 7 % for type III, an 85.3 ± 8 % versus 83.9 ± 7 % for type IV, and a 84.8 ± 8 % versus 81.8 ± 2 % for type V. The best increase in DSC, from the O to the RND CBUS images, was found for the plaque type I (3.86 % increase), with types II and V, following. CONCLUSIONS:In this study, we investigated the impact of CBUS standardization in DL-based carotid plaque type segmentation and showed that indeed normalization of the image resolution and intensity, combined with speckle noise removal, prior to model training and testing, enhances the DL model's performance, across all plaque types. Based on the findings in this study, CBUS images should be standardized when destined for DL-based segmentation tasks, while all plaque types should be considered, as in a plethora of existing relevant studies, uniformly echolucent plaques or heavily calcified plaques with acoustic shadow are notably underrepresented.

The heterogeneity of Multiple Sclerosis (MS) is a challenge for the disease diagnosis and its evolution. In order to monitor the treatment and progression of MS, the segmentation and analysis of brain magnetic resonance imaging (MRI) lesions may offer quantitative evaluation metrics that may be used to compare images across various regions, patients, time points, and institutions. This study targets to provide a comprehensive review of brain MRI studies related to MS disease focusing on lesion segmentation, feature extraction, and Computer-Aided Diagnosis (CAD). Image segmentation methods were categorized regarding their supervision (i.e., supervised, and unsupervised), as well as their deep learning capability. Image analysis focused on feature extraction methods such as texture, structure, and image characterization. Furthermore, to address the challenges and future directions, enable better management of MS disease as well as offer personalized and precision medicine services in clinical praxis, an integrated CAD framework is proposed encompassing 3D reconstruction, registration and visualization, explainable artificial intelligence (AI) and assessment of disease evolution.

BACKGROUND:The rise in life expectancy is associated with an increase in long-term and gradual cognitive decline. Treatment effectiveness is enhanced at the early stage of the disease. Therefore, there is a need to find low-cost and ecological solutions for mass screening of community-dwelling older adults. OBJECTIVE:This work aims to exploit automatic analysis of free speech to identify signs of cognitive function decline. METHODS:A sample of 266 participants older than 65 years were recruited in Italy and Spain and were divided into 3 groups according to their Mini-Mental Status Examination (MMSE) scores. People were asked to tell a story and describe a picture, and voice recordings were used to extract high-level features on different time scales automatically. Based on these features, machine learning algorithms were trained to solve binary and multiclass classification problems by using both mono- and cross-lingual approaches. The algorithms were enriched using Shapley Additive Explanations for model explainability. RESULTS:In the Italian data set, healthy participants (MMSE score≥27) were automatically discriminated from participants with mildly impaired cognitive function (20≤MMSE score≤26) and from those with moderate to severe impairment of cognitive function (11≤MMSE score≤19) with accuracy of 80% and 86%, respectively. Slightly lower performance was achieved in the Spanish and multilanguage data sets. CONCLUSIONS:This work proposes a transparent and unobtrusive assessment method, which might be included in a mobile app for large-scale monitoring of cognitive functionality in older adults. Voice is confirmed to be an important biomarker of cognitive decline due to its noninvasive and easily accessible nature.

Identification of carotid plaque motion alterations, throughout the cardiac cycle (CC), may reveal rupture-prone plaque components and assist doctors in ischemic stroke risk stratification. In this study, we investigated Ultrasound (U/S) carotid plaque motion, focusing on middle-systole (MS), the interval between early systole (ES) and peak systole (PS) expecting that the dynamic forces applied on the plaque at the beginning of systole are more profound from cardiac ES to MS, compared to PS. We deployed a computer tool, previously developed by researchers of the e-health laboratory at Cyprus University of Technology, for carotid plaque motion estimation in B-mode U/S videos. We included 16 carotid U/S videos (8 Asymptomatic, AS; 8 Symptomatic, SY patients) and performed video frame (VF) resolution standardization, manual region of interest (ROI) selection, identification of 5 CCs (ES and MS pairs), and plaque motion spread and magnitude measurement, based on dense optical flow. The Maximum Angular Spread (MAXFW(20), first 20% of the plaque's overall MAXFW; units: 0 degrees to 360 degrees), a measure we created to determine angle differences between motion vectors, in all plaque areas, can assist in classifying the plaque as concordant (low-strain), moderate( m)-discordant (mid-strain) or discordant (high-strain). We identified 1 concordant, 8 m-discordant, and 7 discordant plaques, with MAXFW(20) ranging from 58 +/- 7.45 degrees, 91 +/- 43.9 degrees, 156 +/- 49.2 degrees (mean +/- standard deviation), respectively. As our primary findings support our hypothesis, in the future, we will also derive the stress applied on the carotid plaque, in ES and MS.

The goal of this study is to develop and test an automated integrated speech analysis system for detecting mild cognitive impairment (MCI) and dementia in spontaneous free speech. During the years 2010–2016, speech recordings (N = 2800) were obtained from 200 Greek Cypriots over the age of 65. These were divided into three groups (G 1 , G 2 , and G 3 ) based on the results of their Mini-mental state examination (MMSE): G 1 :95 normal (NOR) individuals with an MMSE greater than 26; G 2 :65 MCI subjects with 20 ≤ MMSE ≤ 26; G 3 :40 dementia subjects with 0 ≤ MMSE < 20. As a result, each speech recording was analyzed for 55 different speech features. The features that could statistically significantly distinguish between the three aforementioned groups were selected using statistical and model multi-classification analysis. Learning-based classifiers were built using the selected features alone or in combination. For each group, statistically significant differences in speech features were detected, which may be used to differentiate the three groups. An overall multi-classification area under the curve (AUC) of 0.92 was attained using only the features identified plus clinical factors. Speech features were extracted, and they were able to discriminate people from the three groups. This study paves the way for the development of an integrated system that uses automatic speech analysis to detect early and progressive signs of cognitive decline (CD) in free speech. In a future study, the proposed method will be developed and integrated into a mobile device.

Monitoring disease evolution in Multiple sclerosis (MS) subjects may aid in decision making for personalizing treatment and disease evolution prediction. We investigate the use of disability progression, using clinical features, the expanded disability status scale (EDSS), and their relationship with texture features and Amplitude Modulation-Frequency Modulation (AM-FM) features extracted from MRI MS detectable lesions for the prognosis of future disability on magnetic resonance imaging (MRI). MS detectable brain lesions from N=38 symptomatic untreated subjects diagnosed with clinically isolated syndrome (CIS), were manually segmented, by an experienced MS neurologist, on transverse T2-weighted (T2W) images obtained from serial brain MRI scans at the baseline (Time0M) and the repeat (Time16-12M) examinations. The subjects were separated into two different groups based on their EDSS: (G1: $1\le $ EDSS $_{\mathrm {2Y}}\le 3.5$ (N=26) and G2: $3.5 < $ EDSS $_{\mathrm {2Y}}\le 8.5$ (N=12) and were monitored over ten years’ time (Time $_{\mathrm {10Y}}$ ). After intensity normalization and image registration, texture and AM-FM features were extracted from all MS lesions at Time0M and Time16-12M. The extracted features were used to develop models that correlated with the disease progression in Time10Y. We found statistically significant differences for features extracted from the two different groups (G1 vs G2 at Time $_{\mathrm {10Y}}$ ) and these might be used to predict the development and or the severity of the MS disease. The best model for classifying G1 vs G2 subjects at Time10Y included information taken from the MS lesion images, texture features and AM-FM features extracted from those MS lesion images (with a correct classification score of %CC=94). The proposed methodology may contribute to additional factors for predicting the development and assessing the severity of the MS disease. However, a larger scale study is needed to establish the application in clinical practice and for computing additional features that may provide information for better and earlier differentiation between normal tissue and MS lesions.

The objective of this study was to implement an explainable artificial intelligence (AI) model with embedded rules to assess Multiple Sclerosis (MS) disease evolution based on brain Magnetic Resonance Imaging (MRI) multi-scale lesion evaluation. Amplitude Modulation-Frequency Modulation (AM-FM) features were extracted from manually segmented brain MS lesions obtained using MRI and were labeled with the Expanded Disability Status Scale (EDSS). Machine learning models were used to classify the MS subjects with a benign course of the disease and subjects with advanced accumulating disability. Rules were extracted from the selected model with high accuracy and then were modified to perform argumentation-based reasoning. It is demonstrated that the proposed explainable AI modeling can distinguish MS subjects and give meaningful information to track the progression of the disease. Future research will examine more subjects and add new feature sets and models.

Multiple Sclerosis (MS) is characterized by complex and heterogeneous nature and as a result, there’s currently no cure. Medications can help control the progression and ease the symptoms of MS. The scientific interest in the field of explainable artificial intelligence (AI) comes to the surface and aims to assist computer-aided diagnostic systems to be established in medical use by providing understandable and transparent information to the experts. The objective of this study was to present different learning methods of explainable AI models in the assessment of MS disease based on clinical data and brain magnetic resonance imaging (MRI) lesion texture features and compare them by focusing on the main findings. The learning methods used machine learning and argumentation theory to differentiate subjects with relapsing-remitting MS (RRMS) from progressive MS (PMS) subjects and provide explanations. The results showed that the different learning methods achieved a high accuracy of 99% and gave similar explanations as they extracted the same set of rules. It is hoped that the proposed methodology could lead to personalized treatment in the management of MS disease.

Transfer learning (TL) reuses knowledge from real-world objects to perform faster and accurate image classification tasks in related content. Multiple studies have shown evaluated deep learning (DL) models in atherosclerotic plaque classification (Asymptomatic, AS, or Symptomatic, SY), using carotid ultrasound (CUS) images, with only a few studies examining TL in this task. In this study, we use TL to classify plaques in CUS longitudinal images, upon image standardization. Overall, 189 images were included (189 patients; 95 SY and 94 AS), which we distributed into training, validation and final evaluation, following the 90–10% data split rule. Our image standardization steps included: image resolution normalization, intensity normalization, and speckle noise removal, followed by cropping of the examples to the plaque region of interest (ROI) and resizing to uniform dimensions. The Xception (Chollet, 2017) and the MobileNet (Howard et al ., 2017) were evaluated in this study, by freezing their backbone architecture (pre-trained on ImageNet) and adding new dense layers, which we trained to classify AS and SY cases. The classification accuracy (CA) of Xception and MobileNet was found at 85% and 75%, respectively. Xception yielded an 81.8% and 88.9% precision for the AS and SY cases, respectively. We also extracted saliency maps (AS or SY), from the best model’s classification layer, during evaluation, to acquire an intuition for plaque areas that play important role in model decision. In the future, we plan to repeat this work with a larger standardized dataset, to also fine-tune layers in model backbones and improve model classification performance.