Journal of Medical Signals & Sensors最新文献

Background: Multiple sclerosis (MS) is an autoimmune disease of the central nervous system which is the main reason of disabilities of young adults. MS occurs when the immune system attacks the central nervous system and destroys the myelin sheaths of neurons. Loss of myelin sheaths results in appearing several lesions in different parts of the brain. The place and amount of lesions are important criteria for determining the level and progression of the disease. These parameters are usually determined manually by an expert which can be time-consuming and inaccurate.

Methods: Considering the effectiveness of artificial intelligence (AI)-based methods in diagnosing and predicting different diseases, and the increasing need for driving new and effective diagnostic methods, this challenge, entitled "Diagnosing MS from magnetic resonance imaging (MRI) Images," has been organized by Isfahan Province Elites Foundation in collaboration with Medical Image and Signal Processing Research Center of Isfahan University of Medical Sciences, as a part of Isfahan AI 2024 event, held in October 2024 in Isfahan, Iran. The challenge has been dedicated to find new AI-based methods for the segmentation and localization of lesions in MRI images of patients with MS. The challenge had three steps, where in the first and second steps, the teams received the train and test datasets, respectively. Finally, the selected teams were invited to the last round of the competition, held in person, and received the last test dataset.

Results: Based on the received results, the best achieved dice score was 0.33, best sensitivity was 0.349, best precision was 0.3, and the lowest centroid distance was 53.025. In addition, the best accuracy for lesion detection in periventricular, deep white matter, juxtacortical, and infratentorial parts of the brain was 80.282%, 74%, 63.492%, and 62.5%, respectively.

Conclusion: Several methods, mostly based on deep learning, have been submitted. The results show that AI has the ability for the segmentation and localization of lesions. However, the received results are still far from the desired accuracy, which shows a need for further improvement and studies in this field.

Background: The use of sedative drugs during various medical procedures is on the rise, necessitating close monitoring of respiratory function throughout the administration process. Continuous auscultation of tracheal sounds is an effective method for monitoring respiratory status. However, it requires constant attention from the operator, which may not always be feasible.

Methods: This concept led to the development of a tracheal sound dataset featuring recordings from 16 patients who underwent cataract surgery at Alzahra Hospital, a university hospital in Isfahan, Iran. To ensure accuracy, the dataset was carefully examined with the assistance of an anesthesiology team, providing precise ground truth annotations for respiratory depression (RD) intervals at a resolution of one second. The Isfahan National Elite Foundation hosted the Isfahan artificial intelligence (AI) 2024 events to advance AI-based detection technologies and offered financial support for five challenges, including the competition for detecting RD from tracheal sounds. Twelve teams from various provinces across Iran participated, utilizing a shared dataset for their evaluations.

Results: The teams that achieved the first through third places were Houshmandsazan, Houshava, and Hoopad, with F1-Scores of 65.18%, 50.44%, and 21.73%, respectively. All participating teams utilized deep learning techniques to detect RD intervals, achieving notable performance, yet opportunities for further improvement remain.

Conclusion: This paper summarizes the performance of these teams, detailing the metrics used to assess their results and the methodologies employed by the top three competitors.

Background: The COVID-19 pandemic has created a critical global situation, causing widespread challenges and numerous fatalities due to severe respiratory complications. Since lung involvement is a key factor in COVID-19 diagnosis and treatment, accurate identification of infected regions in lung images is essential.

Methods: We propose a multiphase segmentation method based on the level set framework to determine lunginvolved areas. The shearlet transform, a high-precision directional multiresolution transform, is employed to guide the gradient flow in the level set formulation. Additionally, the phase stretch transform (PST) is applied to enhance the contrast between infected and healthy regions, improving convergence speed during segmentation.

Results: The proposed algorithm was tested on 500 lung images. The method accurately identified infected areas, enabling precise calculation of the percentage of lung involvement. The use of the shearlet transform also allowed clear delineation of ground-glass opacity boundaries.

Conclusion: The proposed multiphase level set method, enhanced with shearlet and phase stretch transforms, effectively segments COVID-19-infected lung regions. This approach improves segmentation accuracy and computational efficiency, offering a reliable tool for quantitative lung involvement assessment.

Background: Celiac disease (CeD) is a chronic autoimmune condition induced by the consumption of gluten, affecting about 1.4% of the global population. The current diagnostic methods largely rely on serological testing, which may disregard certain biomarkers that are essential for an accurate diagnosis. The objective of the present investigation is to identify significant candidate biomarkers in CeD through using a bioinformatics analysis of microarray data.

Methods: We analyzed three datasets of the Gene Expression Omnibus database (GSE112102, GSE113469, and GSE164883) to conduct a comprehensive bioinformatics approach. We performed a meta-analysis of differentially expressed genes (DEGs), constructed gene ontology and pathway analyses, and developed protein-protein interaction networks to identify and analyze hub genes and their associated miRNAs.

Results: We detected 165 DEGs (79 upregulated and 86 downregulated). Five key hub genes - STAT1, CDC20, perforin-1, CCL2, and MYC were identified as critical regulators involved in controlling both immune system activity and cell cycle progression. Significantly, important miRNAs, including hsa-miR-155-5p, hsa-miR-145-5p, hsa-miR-18a-5p, hsa-miR-34a-5p, hsa-miR-24-3p, and hsa-miR-146a-5p, were seen to have significant interactions with these hub genes. This emphasizes their potential involvement in the pathogenesis of CeD.

Conclusion: The genes identified offer potential as key biomarkers for diagnosing CeD and understanding its molecular mechanisms, creating the path for improved diagnostic and therapeutic strategies.

Background: Accurate dose calculations in radiotherapy are essential, especially in complex anatomical areas such as the nasopharynx, where heterogeneous tissue compositions can greatly influence treatment outcomes. This study assesses the accuracy of the full scatter convolution (FSC) algorithm within the TiGRT treatment planning system by comparing it to the BEAMnrc Monte Carlo (MC) simulation using a head phantom.

Methods: EBT3 film was strategically placed in the nasopharyngeal region to enable direct comparisons between experimental results and those derived from the FSC and MC methods. Various metrics, including the dose difference index, two-dimensional gamma index, and horizontal and vertical dose profiles, were employed for the analysis. The heterogeneous regions were classified into bone, air, and soft-tissue components. For dosimetric evaluation, the irradiated areas were segmented into four regions based on isodose values: Field region (FR), irradiated region (IR), penumbra region (PR), and out-of-FR (OOFR).

Results: The greatest computational discrepancies observed between the FSC algorithm and MC simulations in the air region of the FR were -5.12% ± 1.10% and 1.93% ± 1.45%, respectively. Notable underestimations occurred in the air and soft-tissue regions of the IR, PR, and OOFR when using the FSC algorithm, with a minimum discrepancy of -9.33% ± 5.51% and a maximum of -77.28% ± 8.19%. Conversely, doses calculated for the bone region were overestimated by 53.64% ± 5.65%. In comparison, the MC calculations in the IR region revealed discrepancies of 1.90% ± 1.55% (air), including a maximum underestimation of -8.82% ± 1.18% in the bone area within the PR. The gamma pass rates for different tissue types under local and global modes, using 3%-3 mm gamma criteria, demonstrate that the MC method consistently outperformed the TiGRT method across all tissue types, especially in the air (99.9%) and bone (99.8%) regions.

Conclusions: The findings reveal that the FSC algorithm tends to underestimate doses in soft tissue and air while overestimating doses in bone. In contrast, there was excellent agreement between MC calculations and experimental measurements, highlighting the FSC algorithm's lower consistency.

Background: Alzheimer's disease (AD) is a progressive and irreversible brain disorder, characterized by a gradual decline in cognitive and memory function, with memory loss being one of the most prominent symptoms. Accurate and early diagnosis of AD is essential for effective management and treatment. Structural magnetic resonance imaging (sMRI) and positron emission tomography (PET) are widely utilized neuroimaging modalities for diagnosing AD due to their ability to provide complementary structural and functional insights into brain abnormalities.

Methods: This study introduces a novel computer-aided diagnosis system that integrates sMRI and PET data using Fuzzy Cognitive Maps (FCM) to improve diagnostic accuracy. The research is conducted using the ADNI dataset, where preprocessing of sMRI and PET images is performed using FSL and statistical parametric mapping tools, respectively. In a key innovation, features extracted from both modalities are fused and dimensionality reduction is achieved through an Autoencoder model. The reduced feature set is then classified using FCM, Support Vector Machine, k-Nearest Neighbors, and Multilayer Perceptron.

Results: The FCM-based approach demonstrates superior performance, achieving the highest accuracy of 93.71%, surpassing other classifiers tested.

Conclusions: This study underscores the effectiveness of integrating FCM with multimodal neuroimaging data and highlights its potential for enhancing the early and reliable diagnosis of AD.

Background: Idiopathic ventricular arrhythmia (VA) is among the common cardiac diseases, ranging from benign conditions to those requiring immediate medical intervention. Many VAs originate from the heart's outflow tract (OT). However, this area's complexity and small size, along with other influencing external factors, pose significant challenges to accurate diagnosis. The similarity of the features of VAs on the electrocardiogram (ECG) originating from the right or left side of the OT may lead to misdiagnosis. This study aims to detect the site of origin for VAs originating from the OT, which is important as a key precognition for treatment during catheter ablation.

Methods: We perform this diagnosis using the standard 12-lead ECG and deep learning (DL) techniques without additional equipment. First, inspired by next-generation sequencing in genetics, we created one-dimensional (1D) streams of premature beats from a public dataset of 334 patients. Then, to compare the performance of common 1D DL models, the data were presented to various models, including long short-term memory, gated recurrent unit, and 1D convolutional neural network (1D-CNN).

Results: Experimental results show that the 1D-CNN network achieves the best performance, with an accuracy of 93.4% and an F1-score of 0.9313.

Conclusions: The findings demonstrate the effectiveness of DL in a higher level of applications, specifically in the treatment process, compared to conventional ECG analysis applications based on computerized methods. This represents a promising prospect for use in treatment processes without relying on complex and multifaceted diagnostic methods in the future.

Background: The electroencephalogram (EEG) sonification is an audio portrayal of EEG signals to provide a better understanding of events and brain activity thereupon. This portrayal can be applied to better diagnosis and treatment of some diseases.

Methods: In this study, a new method for EEG sonification is proposed based on extracting musical parameters and note sequences from the dominant frequency ratios and variations in the EEG signal. The ability of different classification structures in extracting musical scales and note sequences is evaluated. A music database has been created to train deep structures which, after extracting the frequency sequence of each piece of music as input, determines the scale label and note sequence in the output. A new algorithm is developed to combine the outputs of the deep structures and create a playable music repertoire.

Results: The findings indicate that the convolutional neural network (CNN) classifier has an accuracy of 93.2% for the classification scales of musical pieces played in different octaves and 92.8% for pieces played in asymmetrical pieces. The convergence of EEG segments with musical scales is also reported for single channel, multi-channel of one person, different individuals, and different databases. The long short-term memory (LSTM) structure selected with an accuracy of 89.6% determines the sequence of notes.

Conclusion: The results show that the proposed CNN determines the appropriate and convergent musical scales with each EEG signal fragment and the LSTM network has a promising performance in converting the dominant frequency variations of EEG signals into note sequences. This demonstrates the good performance of the proposed sonification method.

Background: While metabolic biomarkers are known to play a significant role in the development of ulcerative colitis (UC), the exact causal relationships between them remain uncertain and warrant further investigations. Here we report a bidirectional two-sample Mendelian randomization (MR) study to evaluate causal relationships between 503 blood metabolites and UC.

Methods: We used genome-wide association study (GWAS) data on blood metabolite levels from two separate studies on European individuals (n = 8299 and 24,925). In addition, for UC, we utilized GWAS data from the same ancestry, including 417,932 participants, comprising 5371 UC cases and 412,561 controls. We employed the inverse variance weighted method for our discovery stage of MR analyses. Then, we used other methods, including MR-Egger, weighted median, weighted mode, simple mode, MR-pleiotropy residual sum and outlier, heterogeneity, and pleiotropy tests for sensitivity analyses to further validate our findings and assess the robustness of our results.

Results: Our study suggests that total lipids in small high-density lipoprotein levels (S.HDL.L) are marginal significant positive associated with the development of UC (odds ratio = 1.167, 95% confidence interval: 0.998-1.364, P = 0.051). In addition, UC did not have a statistically significant effect on the metabolites.

Conclusions: Total lipids in S.HDL.L may offer a potential trend as valuable circulating metabolic biomarkers for the screening and prevention of UC in clinical practice. In addition, they could serve as potential candidate molecules for elucidating the mechanisms underlying UC and for identifying suitable drug targets.