Pub Date : 2026-03-01Epub Date: 2025-12-19DOI: 10.1016/j.neuri.2025.100252
Saif M. Balsabti , Rasool M. Al-Gburi , Raid gaib , Ali Mustafa , Shaimaa Khamees Ahmed , Ali Mahmoud Issa , Taha Mahmoud Al-Naimi , Rawan AlSaad , Ali M. Elhenidy
In recent years, the field of medical brain imaging has witnessed remarkable advancements with the integration of artificial intelligence (AI) and deep learning techniques. Traditional unimodal imaging methods, such as MRI and CT, often fall short in providing comprehensive insights into neurological disorders. To address these limitations, multimodal imaging, which combines various imaging modalities like MRI, CT, PET, and SPECT, has emerged as a powerful tool for enhanced diagnosis and treatment planning. This survey presents an in-depth review of the state-of-the-art deep learning models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), used for brain tumor classification, segmentation, forecasting, and object detection. We also explore the potential of hybrid models that integrate machine learning and deep learning approaches. Furthermore, we highlight the significant developments in multimodal brain imaging techniques from 2019 to 2024 and discuss the future research directions needed to advance this field. By synthesizing the latest findings, this survey aims to provide a comprehensive understanding of the current landscape and future possibilities in multimodal medical brain imaging.
{"title":"Advances in deep learning for multimodal brain imaging: A comprehensive survey","authors":"Saif M. Balsabti , Rasool M. Al-Gburi , Raid gaib , Ali Mustafa , Shaimaa Khamees Ahmed , Ali Mahmoud Issa , Taha Mahmoud Al-Naimi , Rawan AlSaad , Ali M. Elhenidy","doi":"10.1016/j.neuri.2025.100252","DOIUrl":"10.1016/j.neuri.2025.100252","url":null,"abstract":"<div><div>In recent years, the field of medical brain imaging has witnessed remarkable advancements with the integration of artificial intelligence (AI) and deep learning techniques. Traditional unimodal imaging methods, such as MRI and CT, often fall short in providing comprehensive insights into neurological disorders. To address these limitations, multimodal imaging, which combines various imaging modalities like MRI, CT, PET, and SPECT, has emerged as a powerful tool for enhanced diagnosis and treatment planning. This survey presents an in-depth review of the state-of-the-art deep learning models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), used for brain tumor classification, segmentation, forecasting, and object detection. We also explore the potential of hybrid models that integrate machine learning and deep learning approaches. Furthermore, we highlight the significant developments in multimodal brain imaging techniques from 2019 to 2024 and discuss the future research directions needed to advance this field. By synthesizing the latest findings, this survey aims to provide a comprehensive understanding of the current landscape and future possibilities in multimodal medical brain imaging.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100252"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924934","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-23DOI: 10.1016/j.neuri.2026.100262
Brock Pluimer , Apeksha Sridhar , Ishtiaq Mawla , Helen Mengxuan Wu , Roshni Lulla , Sarah Hennessy , Patrick Sadil , Rishab Iyer , Eric Ichesco , Anson Kairys , Max Egan , Jonas Kaplan , Richard E. Harris
Careful evaluation of research methodology is fundamental to scientific progress but represents a significant burden on human experts. The complexity of functional MRI (fMRI) methods makes transparent reporting, as suggested by OHBM COBIDAS guidelines, particularly critical. Large Language Models (LLMs) present a potential solution for rapid, scalable methodological assessment. We evaluated three state-of-the-art LLMs (Gemini 2.5 Pro, Claude 4 Sonnet, ChatGPT-o3-pro) against human expert ratings. Fifty fMRI articles (taken from 2016 to 2025) were independently evaluated by ten human experts and three LLMs using an 82-item COBIDAS-based rubric. Human raters demonstrated excellent inter-rater reliability (ICC = 0.801), while LLMs showed poor internal agreement (ICC = 0.254). When comparing total scores across papers, Gemini showed strong positive correlation with human consensus (r = 0.693, p < 0.0001), Claude showed moderate positive correlation (r = 0.394, p = 0.004), while ChatGPT showed negative correlation (r = −0.172, p = 0.233). Gemini maintained high reliability when added to human raters (combined ICC = 0.811), achieving 85.3 % exact agreement and 98.8 % within-1-point agreement. Domain-specific analysis revealed Gemini's consistently high agreement across all six COBIDAS sections (experimental design: 0.915, statistical modeling: 0.880), while ChatGPT and Claude showed weaker, more variable performance. Obvious differences emerged in determining non-applicable items: humans marked 40.5 % as not applicable versus 32.3 % for Gemini, 9.2 % for ChatGPT and 21.1 % for Claude. ChatGPT exhibited extreme score volatility, with papers ranging from 0 to 121 points compared to humans' 44.2–77.7 range. LLM scoring required 1–7 min versus 30–35 min for humans. This proof-of-concept study demonstrates that LLM-assisted methodological evaluation is feasible for complex neuroimaging research and could likely be applied to other research fields.
对研究方法的仔细评估是科学进步的基础，但对人类专家来说却是一项重大负担。功能MRI （fMRI）方法的复杂性使得OHBM COBIDAS指南所建议的透明报告尤为重要。大型语言模型（llm）为快速、可扩展的方法学评估提供了一个潜在的解决方案。我们评估了三个最先进的法学硕士（Gemini 2.5 Pro, Claude 4 Sonnet, chatgpt - 03 - Pro）与人类专家的评级。50篇fMRI文章（取自2016年至2025年）由10位人类专家和3位法学硕士使用基于cobidas的82项标准独立评估。人类评分者表现出优秀的评分者之间的可靠性（ICC = 0.801），而llm表现出较差的内部一致性（ICC = 0.254）。在比较论文总分时，Gemini与人类共识呈强正相关（r = 0.693, p < 0.0001）， Claude与人类共识呈中度正相关（r = 0.394, p = 0.004）， ChatGPT与人类共识呈负相关（r = - 0.172, p = 0.233）。当与人类评分者相结合时，Gemini保持了高可靠性（综合ICC = 0.811），达到85.3%的精确一致性和98.8%的1点以内一致性。特定领域的分析显示，Gemini在所有六个COBIDAS部分（实验设计：0.915，统计建模：0.880）的一致性始终很高，而ChatGPT和Claude表现出更弱、更多变的表现。在确定不适用的项目上出现了明显的差异：人类将40.5%标记为不适用，而双子座为32.3%，ChatGPT为9.2%，克劳德为21.1%。ChatGPT表现出极端的得分波动，论文的得分范围在0到121分之间，而人类的得分范围在44.2到77.7分之间。LLM评分需要1-7分钟，而人类评分需要30-35分钟。这项概念验证研究表明，法学硕士辅助的方法评估对于复杂的神经影像学研究是可行的，并且可能应用于其他研究领域。
{"title":"A proof-of-concept study on the use of large language models for assessing research methodology in neuroimaging","authors":"Brock Pluimer , Apeksha Sridhar , Ishtiaq Mawla , Helen Mengxuan Wu , Roshni Lulla , Sarah Hennessy , Patrick Sadil , Rishab Iyer , Eric Ichesco , Anson Kairys , Max Egan , Jonas Kaplan , Richard E. Harris","doi":"10.1016/j.neuri.2026.100262","DOIUrl":"10.1016/j.neuri.2026.100262","url":null,"abstract":"<div><div>Careful evaluation of research methodology is fundamental to scientific progress but represents a significant burden on human experts. The complexity of functional MRI (fMRI) methods makes transparent reporting, as suggested by OHBM COBIDAS guidelines, particularly critical. Large Language Models (LLMs) present a potential solution for rapid, scalable methodological assessment. We evaluated three state-of-the-art LLMs (Gemini 2.5 Pro, Claude 4 Sonnet, ChatGPT-o3-pro) against human expert ratings. Fifty fMRI articles (taken from 2016 to 2025) were independently evaluated by ten human experts and three LLMs using an 82-item COBIDAS-based rubric. Human raters demonstrated excellent inter-rater reliability (ICC = 0.801), while LLMs showed poor internal agreement (ICC = 0.254). When comparing total scores across papers, Gemini showed strong positive correlation with human consensus (r = 0.693, p < 0.0001), Claude showed moderate positive correlation (r = 0.394, p = 0.004), while ChatGPT showed negative correlation (r = −0.172, p = 0.233). Gemini maintained high reliability when added to human raters (combined ICC = 0.811), achieving 85.3 % exact agreement and 98.8 % within-1-point agreement. Domain-specific analysis revealed Gemini's consistently high agreement across all six COBIDAS sections (experimental design: 0.915, statistical modeling: 0.880), while ChatGPT and Claude showed weaker, more variable performance. Obvious differences emerged in determining non-applicable items: humans marked 40.5 % as not applicable versus 32.3 % for Gemini, 9.2 % for ChatGPT and 21.1 % for Claude. ChatGPT exhibited extreme score volatility, with papers ranging from 0 to 121 points compared to humans' 44.2–77.7 range. LLM scoring required 1–7 min versus 30–35 min for humans. This proof-of-concept study demonstrates that LLM-assisted methodological evaluation is feasible for complex neuroimaging research and could likely be applied to other research fields.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100262"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-13DOI: 10.1016/j.neuri.2026.100261
Jiantao Shen , Sung-Min Jun , Samantha J. Holdsworth , Gonzalo Maso Talou , Jason A. Correia , Hamid Abbasi
Accurate and automated meningioma segmentation remains a biomedical engineering challenge, particularly when relying on single-modality MRI data. We evaluate SegResNet, a U-Net-based deep learning architecture, for meningioma segmentation using 817 T1 contrast-enhanced (T1CE) magnetic resonance imaging (MRI) images from 282 patients across Auckland, New Zealand. We investigate the effect of incorporating additional images from the 2023 Brain Tumor Segmentation (BraTS) meningioma challenge during training on model performance. The baseline model trained solely on the Auckland dataset achieved 75.67 % mean Dice. Incorporating an additional 200 and 400 BraTS images improved segmentation performance to 77.89 % and 76.73 %, respectively. A separate experiment involving pre-training on BraTS data followed by fine-tuning on Auckland data achieved 75.90 % Dice. Our results suggest that while leveraging external datasets can enhance model robustness, the extent of improvement depends on dataset heterogeneity and alignment with the target domain.
Analysis of a subset of images unaffected by skull-stripping artifacts indicated notably higher segmentation accuracy (up to 84.02 % Dice), highlighting the influence of preprocessing on performance. Evaluations using the 2023 and 2024 BraTS lesion-wise metrics demonstrated the importance of context-appropriate metric selection. Our findings highlight the adaptability of SegResNet to a single-modality T1CE – a widely available sequence in standard clinical protocols – clinical dataset and emphasize how public data integration, careful preprocessing, and task-aligned evaluation can support robust segmentation models for diverse and resource-constrained environments.
{"title":"Evaluating SegResNet for single-modality meningioma segmentation on T1 contrast-enhanced MRI on a New Zealand clinical cohort","authors":"Jiantao Shen , Sung-Min Jun , Samantha J. Holdsworth , Gonzalo Maso Talou , Jason A. Correia , Hamid Abbasi","doi":"10.1016/j.neuri.2026.100261","DOIUrl":"10.1016/j.neuri.2026.100261","url":null,"abstract":"<div><div>Accurate and automated meningioma segmentation remains a biomedical engineering challenge, particularly when relying on single-modality MRI data. We evaluate SegResNet, a U-Net-based deep learning architecture, for meningioma segmentation using 817 T1 contrast-enhanced (T1CE) magnetic resonance imaging (MRI) images from 282 patients across Auckland, New Zealand. We investigate the effect of incorporating additional images from the 2023 Brain Tumor Segmentation (BraTS) meningioma challenge during training on model performance. The baseline model trained solely on the Auckland dataset achieved 75.67 % mean Dice. Incorporating an additional 200 and 400 BraTS images improved segmentation performance to 77.89 % and 76.73 %, respectively. A separate experiment involving pre-training on BraTS data followed by fine-tuning on Auckland data achieved 75.90 % Dice. Our results suggest that while leveraging external datasets can enhance model robustness, the extent of improvement depends on dataset heterogeneity and alignment with the target domain.</div><div>Analysis of a subset of images unaffected by skull-stripping artifacts indicated notably higher segmentation accuracy (up to 84.02 % Dice), highlighting the influence of preprocessing on performance. Evaluations using the 2023 and 2024 BraTS lesion-wise metrics demonstrated the importance of context-appropriate metric selection. Our findings highlight the adaptability of SegResNet to a single-modality T1CE – a widely available sequence in standard clinical protocols – clinical dataset and emphasize how public data integration, careful preprocessing, and task-aligned evaluation can support robust segmentation models for diverse and resource-constrained environments.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100261"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037174","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-06DOI: 10.1016/j.neuri.2026.100265
Fabio Musio , Norman Juchler , Kaiyuan Yang , Suprosanna Shit , Chinmay Prabhakar , Bjoern Menze , Sven Hirsch
The Circle of Willis (CoW) is a critical network of brain arteries, often implicated in cerebrovascular pathologies. Voxel-level segmentation is an important first step toward automated CoW assessment, but full quantitative analysis requires centerline representations. However, conventional skeletonization techniques often struggle to extract reliable centerlines due to the CoW's complex geometry, and publicly available centerline datasets remain scarce. To address these challenges, we used a thinning-based skeletonization algorithm to extract and curate centerline graphs and morphometric features from the TopCoW dataset, which includes 200 stroke patients imaged with magnetic resonance angiography (MRA) and computed tomography angiography (CTA). The curated graphs were used to develop a baseline algorithm for centerline and feature extraction, combining U-Net-based skeletonization with A∗ graph connection. Performance was evaluated on a held-out test set, focusing on anatomical accuracy and feature robustness. Further, we used the extracted features to predict the frequency of fetal-type PCA, confirm theoretical bifurcation optimality relations, and detect subtle modality differences. The baseline algorithm consistently reconstructed graph topology with high accuracy (F1 = 1), and average node distance between reference and predicted graphs was below one voxel. Features such as segment radius, length, and bifurcation ratios showed strong robustness, with median relative errors below 5% and Pearson correlations above 0.95. Our results demonstrate the utility of learning-based skeletonization for anatomically plausible centerline extraction. We emphasize the importance of going beyond voxel-level metrics by evaluating anatomical accuracy and feature robustness. The dataset and baseline algorithm have been released to support further research.
{"title":"Circle of Willis centerline graphs: A dataset and baseline algorithm","authors":"Fabio Musio , Norman Juchler , Kaiyuan Yang , Suprosanna Shit , Chinmay Prabhakar , Bjoern Menze , Sven Hirsch","doi":"10.1016/j.neuri.2026.100265","DOIUrl":"10.1016/j.neuri.2026.100265","url":null,"abstract":"<div><div>The Circle of Willis (CoW) is a critical network of brain arteries, often implicated in cerebrovascular pathologies. Voxel-level segmentation is an important first step toward automated CoW assessment, but full quantitative analysis requires centerline representations. However, conventional skeletonization techniques often struggle to extract reliable centerlines due to the CoW's complex geometry, and publicly available centerline datasets remain scarce. To address these challenges, we used a thinning-based skeletonization algorithm to extract and curate centerline graphs and morphometric features from the TopCoW dataset, which includes 200 stroke patients imaged with magnetic resonance angiography (MRA) and computed tomography angiography (CTA). The curated graphs were used to develop a baseline algorithm for centerline and feature extraction, combining U-Net-based skeletonization with A∗ graph connection. Performance was evaluated on a held-out test set, focusing on anatomical accuracy and feature robustness. Further, we used the extracted features to predict the frequency of fetal-type PCA, confirm theoretical bifurcation optimality relations, and detect subtle modality differences. The baseline algorithm consistently reconstructed graph topology with high accuracy (F1 = 1), and average node distance between reference and predicted graphs was below one voxel. Features such as segment radius, length, and bifurcation ratios showed strong robustness, with median relative errors below 5% and Pearson correlations above 0.95. Our results demonstrate the utility of learning-based skeletonization for anatomically plausible centerline extraction. We emphasize the importance of going beyond voxel-level metrics by evaluating anatomical accuracy and feature robustness. The dataset and baseline algorithm have been released to support further research.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100265"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In current medical practice, patients undergoing treatment for depression typically must wait four to six weeks before clinicians can assess their response to medication, due to the delayed onset of noticeable effects from antidepressants. Identifying treatment response at an earlier stage is of great importance, as it can reduce both the emotional and economic burden associated with prolonged treatment. We present a novel Motif Discovery Framework (MDF) that extracts dynamic features from EEG time series data to distinguish between treatment responders and non-responders in depression. Our findings show that MDF can predict treatment response with high precision as early as the 7th day of treatment, significantly reducing the waiting time for patients. Furthermore, we demonstrate that MDF generalizes well to classification tasks in other psychiatric conditions, including schizophrenia, Alzheimer’s disease, and dementia. Overall, our experiments show that MDF outperforms relevant benchmarks. The high precision of our classification framework underscores the potential of EEG dynamic properties-represented as motifs-to support clinical decision-making and ultimately enhance patient quality of life.
{"title":"EEG-based classification in psychiatry using motif discovery","authors":"Melanija Kraljevska , Kateřina Hlaváčková-Schindler , Lukas Miklautz , Claudia Plant","doi":"10.1016/j.neuri.2025.100242","DOIUrl":"10.1016/j.neuri.2025.100242","url":null,"abstract":"<div><div>In current medical practice, patients undergoing treatment for depression typically must wait four to six weeks before clinicians can assess their response to medication, due to the delayed onset of noticeable effects from antidepressants. Identifying treatment response at an earlier stage is of great importance, as it can reduce both the emotional and economic burden associated with prolonged treatment. We present a novel Motif Discovery Framework (MDF) that extracts dynamic features from EEG time series data to distinguish between treatment responders and non-responders in depression. Our findings show that MDF can predict treatment response with high precision as early as the 7th day of treatment, significantly reducing the waiting time for patients. Furthermore, we demonstrate that MDF generalizes well to classification tasks in other psychiatric conditions, including schizophrenia, Alzheimer’s disease, and dementia. Overall, our experiments show that MDF outperforms relevant benchmarks. The high precision of our classification framework underscores the potential of EEG dynamic properties-represented as motifs-to support clinical decision-making and ultimately enhance patient quality of life.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100242"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145618463","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2025-12-18DOI: 10.1016/j.neuri.2025.100253
Abdalla Nabil Elsharkawy , Nourhan Zayed
Intraoperative neuromonitoring (IONM) plays a critical role in preserving nerve function during high-risk surgeries through real-time monitoring of electromyographic (EMG) activity. Routine EMG analysis, in real-time, is complex and prone to variability. This work presents an end-to-end deep learning-based framework for accurate EMG signal classification of the nerve status using the discrete wavelet transform (DWT) mathematical technique. Four state-of-the-art deep learning architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a CNN-LSTM ensemble, and a Transformer model, were tested with various Daubechies wavelet families (db1–db6) and window sizes (50–500 samples). The Transformer model performed superiorly in classification, achieving an outstanding accuracy of 98.13 %, an F1-score of 98.14 %, and a recall of 97.50 % using db1 and a 400-sample window. The results summed up that the use of wavelet-based time-frequency decomposition has a significant influence on enhancing classification performance, especially when utilized with deep learning models.
{"title":"Exploring the effects of wavelet types and windowing on EMG-based IONM through deep learning architectures","authors":"Abdalla Nabil Elsharkawy , Nourhan Zayed","doi":"10.1016/j.neuri.2025.100253","DOIUrl":"10.1016/j.neuri.2025.100253","url":null,"abstract":"<div><div>Intraoperative neuromonitoring (IONM) plays a critical role in preserving nerve function during high-risk surgeries through real-time monitoring of electromyographic (EMG) activity. Routine EMG analysis, in real-time, is complex and prone to variability. This work presents an end-to-end deep learning-based framework for accurate EMG signal classification of the nerve status using the discrete wavelet transform (DWT) mathematical technique. Four state-of-the-art deep learning architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a CNN-LSTM ensemble, and a Transformer model, were tested with various Daubechies wavelet families (db1–db6) and window sizes (50–500 samples). The Transformer model performed superiorly in classification, achieving an outstanding accuracy of 98.13 %, an F1-score of 98.14 %, and a recall of 97.50 % using db1 and a 400-sample window. The results summed up that the use of wavelet-based time-frequency decomposition has a significant influence on enhancing classification performance, especially when utilized with deep learning models.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100253"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924926","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2025-10-28DOI: 10.1016/j.neuri.2025.100238
Esteban A. Alarcón-Braga , Samuel Gruffaz , Cécile Delagarde , Axel Roques , Jean-Clément Riff , Laurent Oudre , Clément Dubost
Existing methods to detect depth of sedation do not fully adjust to the characteristics of the ICU population. The aim of this study is to evaluate the performance of a two-channel EEG in predicting the depth of sedation in ICU patients. The electroencephalographic signal of 21 patients admitted to the ICU were analyzed, and EEG features were calculated. These served as inputs in 2 machine learning models: Random Forest Classifier (RFC) and Support Vector Machine (SVM). The depth of sedation was assessed using the Richmond Agitation-Sedation Scale (RASS). Patients with RASS scores of −4/−5 were classified as “Deeply Sedated”, otherwise they were classified as “Not Deeply Sedated”. In the general models, all EEG features were used, after which sequential feature selection was conducted to improve performance and reduce the number of variables (reduced models). The general models showed a moderate ability to discriminate between sedation categories (RFC: average , SVM: average ). This ability was improved in the reduced models (RFC: average , SVM: average ). It was observed that decreasing the number of features in the reduced SVM model from 6 to 3 features could achieve a similar performance while simplifying the model (SVM: average ). An exploratory analysis showed that the individual feature with the best performance was Beta Power–EEG2. Overall, the 2-channel EEG has a moderate power to discriminate between different states of sedation and may not be useful in this purpose if used as a single predictor.
{"title":"Detecting the depth of sedation in the intensive care unit using a 2-channel electroencephalogram: An analysis with 2 machine learning models","authors":"Esteban A. Alarcón-Braga , Samuel Gruffaz , Cécile Delagarde , Axel Roques , Jean-Clément Riff , Laurent Oudre , Clément Dubost","doi":"10.1016/j.neuri.2025.100238","DOIUrl":"10.1016/j.neuri.2025.100238","url":null,"abstract":"<div><div>Existing methods to detect depth of sedation do not fully adjust to the characteristics of the ICU population. The aim of this study is to evaluate the performance of a two-channel EEG in predicting the depth of sedation in ICU patients. The electroencephalographic signal of 21 patients admitted to the ICU were analyzed, and EEG features were calculated. These served as inputs in 2 machine learning models: Random Forest Classifier (RFC) and Support Vector Machine (SVM). The depth of sedation was assessed using the Richmond Agitation-Sedation Scale (RASS). Patients with RASS scores of −4/−5 were classified as “Deeply Sedated”, otherwise they were classified as “Not Deeply Sedated”. In the general models, all EEG features were used, after which sequential feature selection was conducted to improve performance and reduce the number of variables (reduced models). The general models showed a moderate ability to discriminate between sedation categories (RFC: average <span><math><mtext>F1-score</mtext><mo>=</mo><mn>0.60</mn></math></span>, SVM: average <span><math><mtext>F1-score</mtext><mo>=</mo><mn>0.59</mn></math></span>). This ability was improved in the reduced models (RFC: average <span><math><mtext>F1-score</mtext><mo>=</mo><mn>0.65</mn></math></span>, SVM: average <span><math><mtext>F1-score</mtext><mo>=</mo><mn>0.72</mn></math></span>). It was observed that decreasing the number of features in the reduced SVM model from 6 to 3 features could achieve a similar performance while simplifying the model (SVM: average <span><math><mtext>F1-score</mtext><mo>=</mo><mn>0.72</mn></math></span>). An exploratory analysis showed that the individual feature with the best performance was Beta Power–EEG2. Overall, the 2-channel EEG has a moderate power to discriminate between different states of sedation and may not be useful in this purpose if used as a single predictor.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100238"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977036","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2025-12-19DOI: 10.1016/j.neuri.2025.100249
Selorm Adablanu , Utpal Barman , Dulumani Das
Optimization algorithms are pivotal in training deep learning (DL) models for medical imaging, determining how efficiently models learn, generalize, and perform across modalities. This systematic review analyzed 69 peer-reviewed studies (2010–2025) on optimizer performance in classification, segmentation, and object detection tasks using MRI, CT, X-ray, ultrasound, histopathology, and ECG data, following PRISMA 2020 guidelines. Adaptive optimizers such as Adam and its variants were most common, offering rapid convergence in CNN-based classification, whereas SGD and momentum-based methods yielded stronger generalization in large-scale segmentation. Emerging techniques—Sharpness-Aware Minimization (SAM), Ranger, and AdamW—improved robustness under domain shift or noisy conditions. Hybrid and metaheuristic optimizers provided marginal accuracy gains but at higher computational cost. Common limitations included inconsistent hyperparameter reporting, limited external validation, and dataset bias toward North American cohorts. Optimizer effectiveness was found to be task- and architecture-dependent: adaptive methods suit small or noisy datasets, while momentum-based and hybrid methods enhance generalization for complex imaging. Future studies should emphasize standardized evaluation, transparent reporting, and diverse data to enable equitable and reproducible deployment of medical AI.
{"title":"15 Years of optimizers in medical deep learning: A systematic review","authors":"Selorm Adablanu , Utpal Barman , Dulumani Das","doi":"10.1016/j.neuri.2025.100249","DOIUrl":"10.1016/j.neuri.2025.100249","url":null,"abstract":"<div><div>Optimization algorithms are pivotal in training deep learning (DL) models for medical imaging, determining how efficiently models learn, generalize, and perform across modalities. This systematic review analyzed 69 peer-reviewed studies (2010–2025) on optimizer performance in classification, segmentation, and object detection tasks using MRI, CT, X-ray, ultrasound, histopathology, and ECG data, following PRISMA 2020 guidelines. Adaptive optimizers such as Adam and its variants were most common, offering rapid convergence in CNN-based classification, whereas SGD and momentum-based methods yielded stronger generalization in large-scale segmentation. Emerging techniques—Sharpness-Aware Minimization (SAM), Ranger, and AdamW—improved robustness under domain shift or noisy conditions. Hybrid and metaheuristic optimizers provided marginal accuracy gains but at higher computational cost. Common limitations included inconsistent hyperparameter reporting, limited external validation, and dataset bias toward North American cohorts. Optimizer effectiveness was found to be task- and architecture-dependent: adaptive methods suit small or noisy datasets, while momentum-based and hybrid methods enhance generalization for complex imaging. Future studies should emphasize standardized evaluation, transparent reporting, and diverse data to enable equitable and reproducible deployment of medical AI.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100249"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-15DOI: 10.1016/j.neuri.2026.100258
Tsubasa Sasaki , Yoshiyuki Hirano
Background
Many studies on resting-state functional connectivity (FC) in major depressive disorder (MDD) have investigated FC as a biomarker of disease pathogenesis. However, few studies have examined conditional dependencies among FC, clinical status, and demographic variables. Considering such dependencies allows the identification of direct relationships obscured by spurious correlations.
Aim
This study aimed to examine the neural mechanisms of MDD and propose a structural relationship between FC and MDD, focusing on sulcal regions.
Methods
Using a large dataset of 431 healthy controls and 235 MDD patients with MDD, we combined partial least squares (PLS)-based feature extraction with logistic regression and light gradient boosting machine (LightGBM) models for diagnostic classification, followed by Bayesian network (BN) analysis employing a directed acyclic graph.
Results
The classification models demonstrated moderate accuracy (logistic regression: area under the curve [AUC] = 0.735; LightGBM: AUC = 0.710). Structure learning with the Max–Min Hill-Climbing algorithm revealed direct edges from the MDD diagnosis to variables derived from the BDI and PLS components, but no direct parent nodes of MDD were identified. Intervention simulation showed that the MDD diagnosis significantly reduced FC in the default mode network (DMN), dorsal attention network, and between subcortical structures and cortex.
Conclusion
MDD diagnosis is associated with disease-specific disruptions not only in the DMN but also across multiple networks, underscoring the need to consider widespread network dysfunction in the pathophysiology of MDD. Future longitudinal and interventional research is required to clarify the causal relationships between the diagnosis and brain function.
{"title":"A study on the potential relationship between the diagnosis and functional connectivity in the brain in major depressive disorder","authors":"Tsubasa Sasaki , Yoshiyuki Hirano","doi":"10.1016/j.neuri.2026.100258","DOIUrl":"10.1016/j.neuri.2026.100258","url":null,"abstract":"<div><h3>Background</h3><div>Many studies on resting-state functional connectivity (FC) in major depressive disorder (MDD) have investigated FC as a biomarker of disease pathogenesis. However, few studies have examined conditional dependencies among FC, clinical status, and demographic variables. Considering such dependencies allows the identification of direct relationships obscured by spurious correlations.</div></div><div><h3>Aim</h3><div>This study aimed to examine the neural mechanisms of MDD and propose a structural relationship between FC and MDD, focusing on sulcal regions.</div></div><div><h3>Methods</h3><div>Using a large dataset of 431 healthy controls and 235 MDD patients with MDD, we combined partial least squares (PLS)-based feature extraction with logistic regression and light gradient boosting machine (LightGBM) models for diagnostic classification, followed by Bayesian network (BN) analysis employing a directed acyclic graph.</div></div><div><h3>Results</h3><div>The classification models demonstrated moderate accuracy (logistic regression: area under the curve [AUC] = 0.735; LightGBM: AUC = 0.710). Structure learning with the Max–Min Hill-Climbing algorithm revealed direct edges from the MDD diagnosis to variables derived from the BDI and PLS components, but no direct parent nodes of MDD were identified. Intervention simulation showed that the MDD diagnosis significantly reduced FC in the default mode network (DMN), dorsal attention network, and between subcortical structures and cortex.</div></div><div><h3>Conclusion</h3><div>MDD diagnosis is associated with disease-specific disruptions not only in the DMN but also across multiple networks, underscoring the need to consider widespread network dysfunction in the pathophysiology of MDD. Future longitudinal and interventional research is required to clarify the causal relationships between the diagnosis and brain function.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100258"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-14DOI: 10.1016/j.neuri.2026.100260
Frank te Nijenhuis , Matthijs van der Sluijs , Pieter Jan van Doormaal , Wim van Zwam , Jeannette Hofmeijer , Xucong Zhang , Sandra Cornelissen , Danny Ruijters , Ruisheng Su , Theo van Walsum
In acute ischemic stroke, large vessel occlusions of the anterior circulation are increasingly treated with endovascular therapy (EVT). The efficacy of this therapy depends on adequate treatment selection. Treatment decisions can be based on predictions of functional outcome. Most existing studies predict functional outcomes using clinical parameters. We set out to study functional outcome prediction performance by integrating imaging in a multimodal setting. Using a multi-center dataset containing 2927 patients, we compare the functional outcome prediction performances of clinical baseline models, including the clinically validated MR PREDICTS decision tool, image-based models with deep learning networks, and a multimodal approach combining clinical and imaging information. The predicted outcome measure is dichotomized modified Rankin Scale score 90 days after EVT. We perform sanity checks, hyperparameter optimization, and comparisons of effectiveness of using CTA, NCCT, or both images as input. Our experiments show that information extracted from CTA or NCCT images does not significantly improve the performance, as quantified using AUC, of functional outcome prediction methods compared to a baseline model. The multimodal approach may replace radiologically derived biomarkers, as its performance is non-inferior.
{"title":"Integrating cross-sectional imaging data into functional outcome prediction models for acute ischemic stroke of the anterior circulation","authors":"Frank te Nijenhuis , Matthijs van der Sluijs , Pieter Jan van Doormaal , Wim van Zwam , Jeannette Hofmeijer , Xucong Zhang , Sandra Cornelissen , Danny Ruijters , Ruisheng Su , Theo van Walsum","doi":"10.1016/j.neuri.2026.100260","DOIUrl":"10.1016/j.neuri.2026.100260","url":null,"abstract":"<div><div>In acute ischemic stroke, large vessel occlusions of the anterior circulation are increasingly treated with endovascular therapy (EVT). The efficacy of this therapy depends on adequate treatment selection. Treatment decisions can be based on predictions of functional outcome. Most existing studies predict functional outcomes using clinical parameters. We set out to study functional outcome prediction performance by integrating imaging in a multimodal setting. Using a multi-center dataset containing 2927 patients, we compare the functional outcome prediction performances of clinical baseline models, including the clinically validated MR PREDICTS decision tool, image-based models with deep learning networks, and a multimodal approach combining clinical and imaging information. The predicted outcome measure is dichotomized modified Rankin Scale score 90 days after EVT. We perform sanity checks, hyperparameter optimization, and comparisons of effectiveness of using CTA, NCCT, or both images as input. Our experiments show that information extracted from CTA or NCCT images does not significantly improve the performance, as quantified using AUC, of functional outcome prediction methods compared to a baseline model. The multimodal approach may replace radiologically derived biomarkers, as its performance is non-inferior.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100260"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号