Informatics in Medicine Unlocked最新文献

{"title":"Enhanced ROI guided deep learning model for Alzheimer’s detection using 3D MRI images","authors":"Israt Jahan Khan ,&nbsp;Md. Fahim Bin Amin ,&nbsp;Md. Delwar Shahadat Deepu ,&nbsp;Hazera Khatun Hira ,&nbsp;Asif Mahmud ,&nbsp;Anas Mashad Chowdhury ,&nbsp;Salekul Islam ,&nbsp;Md. Saddam Hossain Mukta ,&nbsp;Swakkhar Shatabda ,&nbsp;Alzheimer’s Disease Neuroimaging Initiative","doi":"10.1016/j.imu.2025.101650","DOIUrl":"10.1016/j.imu.2025.101650","url":null,"abstract":"<div><div>Alzheimer’s disease is an incurable condition that predominantly affects the human brain, leading to the shrinkage of various brain regions and the disruption of neuronal connections. Current state-of-the-art methods for detecting Alzheimer’s disease using 3D MRI images are resource-intensive and time-consuming. In this paper, we propose a Regions of Interest (ROI)-guided detection paradigm to address these challenges. We employ a 3D ResNet integrated with a Convolutional Block Attention Module (CBAM), demonstrating that emphasising ROIs in brain imaging can substantially reduce both computational expenditure and training time. Our model exhibits robust performance in discriminating Alzheimer’s disease from mild cognitive impairment, achieving an accuracy of 88% across the entire brain and 92% within targeted ROIs on the ADNI dataset. The accuracy on the OASIS dataset is even higher, reaching 98% for all regions and 98.33% for the ROIs. When distinguishing Alzheimer’s disease from cognitively normal individuals, the accuracy improves further, achieving 93.33% for the ROIs on the ADNI dataset and 97.8% on the OASIS dataset. In differentiating cognitively normal individuals from those with mild cognitive impairment, the model attains an accuracy of 88.2% for the ROIs on the ADNI dataset and 98.6% on the OASIS dataset. These findings highlight a notable enhancement in detection accuracy through the utilisation of fewer, yet more salient brain regions, underscoring the efficacy of our ROI-guided approach.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"56 ","pages":"Article 101650"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124751","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}

{"title":"A lightweight classification system for the early detection of diabetic retinopathy","authors":"Ashim Chakraborty,&nbsp;George Wilson,&nbsp;Cristina Luca","doi":"10.1016/j.imu.2025.101655","DOIUrl":"10.1016/j.imu.2025.101655","url":null,"abstract":"<div><div>The eye disease known as Diabetic Retinopathy is one of the leading causes of permanent blindness in people of working age worldwide. Early identification is crucial for the treatment and management of the condition and this study presents a trustworthy approach for identifying the early stages of the disease from fundus images. A comparative analysis of a supervised machine learning algorithm and manual classification conducted by qualified optometrists is used to evaluate the work. Diabetic Retinopathy features such as Hard Exudates, Microaneurysms and Blood Vessels are extracted from the retinal images by a number of feature extraction methods. The performance and robustness of the proposed novel system are assessed using confusion matrix data and AUC-ROC curves. The findings demonstrate the validity of the decision-based system for the early detection of diabetic retinopathy, with the potential to be deployed on a portable screening system that can be used by people living in remote areas of the world.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"57 ","pages":"Article 101655"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144230269","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}

{"title":"WAE-DTI: Ensemble-based architecture for drug–target interaction prediction using descriptors and embeddings","authors":"Tariq Sha’ban,&nbsp;Ahmad M. Mustafa,&nbsp;Mostafa Z. Ali","doi":"10.1016/j.imu.2024.101604","DOIUrl":"10.1016/j.imu.2024.101604","url":null,"abstract":"<div><div>Drug Target Interaction (DTI) prediction is one of the main challenges in the pharmaceutical and drug discovery domains due to its high costs, time-consuming nature, and complexity of manual experiments required to evaluate interactions between large numbers of drugs and targets. In addition, a single drug can bind to multiple targets. In contrast, a single target can also bind to a number of drugs; this complicates the DTI task. Existing silico models often struggle with these challenges, particularly in managing diverse datasets. To address these issues, we introduce the Weighted Average Ensemble Drug–Target Interaction (WAE-DTI) model. Our approach integrates several descriptors and fingerprint representations to enhance prediction accuracy and generalization, namely atom pair fingerprint, Avalon, MACCS, MH, Morgan, RDKit, SEC, topological torsion, and LDP for drug representation, and ESM-2 for target representation. WAE-DTI employs a weighted average ensemble technique to handle diverse datasets effectively. The model demonstrates significant improvements over state-of-the-art methods, achieving an average mean squared error of 0.190 ±0.001 on the Davis dataset, 0.127 ±0.001 on Kiba, 0.143 ±0.001 on DTC, 0.284 ±0.004 on Metz, 0.308 ±0.001 on ToxCast, and 0.934 ±0.004 on STITCH. As for the classification task, WAE-DTI outperforms existing models with an AUPRC of 0.943 ±0.001 on BioSNAP, 0.474 ±0.011 on Davis, and 0.707 ±0.005 on BindingDB. Our code is publicly available at <span><span>¹</span></span>.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"52 ","pages":"Article 101604"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143178371","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Do transformers generalise better than bespoke tools for anonymisation?","authors":"Roman Klapaukh ,&nbsp;Carol El-Hayek ,&nbsp;Douglas IR Boyle","doi":"10.1016/j.imu.2024.101607","DOIUrl":"10.1016/j.imu.2024.101607","url":null,"abstract":"<div><div>Free-text fields in clinical records contain information that may not show up in the structured health record. Automated anonymisation tools can lower the bar to using this data at scale. However, existing anonymisation tools do not always perform as well as expected when used outside of their country and domain of origin. We ran three US tertiary care targeting transformer models on 300 Australian general practice notes, and showed that they generalise better than purpose built tools.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"52 ","pages":"Article 101607"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143178368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Brain tumor classification using a hybrid ensemble of Xception and parallel deep CNN models","authors":"Seoyoung Yoon","doi":"10.1016/j.imu.2025.101629","DOIUrl":"10.1016/j.imu.2025.101629","url":null,"abstract":"<div><h3>Objective</h3><div>Accurate classification of brain tumors is essential for effective diagnosis and treatment planning. The purpose of this study is to develop and evaluate a hybrid ensemble brain tumor classification method to leverage the strengths of two different architectures for improving the accuracy, robustness, and reliability of brain tumor classification.</div></div><div><h3>Methodology</h3><div>This study introduces a novel and innovative classifier that concatenates the Xception convolutional neural network (CNN) with kernel size of (3,3) and a parallel deep CNN (PDCNN) with kernel size of (5,5) and (12,12) to classify brain tumor images from the Kaggle dataset into four categories: meningioma, glioma, pituitary, and no tumor.</div></div><div><h3>Results</h3><div>The Xception model alone achieved a classification accuracy of 98.26 %, while the PDCNN model achieved 94.85 % on the same dataset. By concatenating these two models, the proposed hybrid ensemble approach enhanced overall classification accuracy to 99.09 %. In comparison with state-of-the-art models, VGG19 achieved an accuracy of 94.69 %, while ResNet152V2 achieved 96.27 % on the same dataset. The proposed hybrid ensemble model with Xception and PDCNN consistently outperformed both VGG19 and ResNet152V2.</div></div><div><h3>Conclusion</h3><div>This synergy of concatenating the Xception and PDCNN architectures demonstrates the innovativeness and effectiveness of leveraging complementary strengths in feature extraction and classification, leading to enhanced performance in brain tumor detection. The results highlight the potential of ensemble deep learning models in advancing automated medical image analysis and improving clinical outcomes.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"54 ","pages":"Article 101629"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143600465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Structural modification of Naproxen; physicochemical, spectral, medicinal, and pharmacological evaluation","authors":"Md Omor Farque ,&nbsp;Rahat Moinul Islam ,&nbsp;Md Ferdous Rahman Joni ,&nbsp;Mimona Akter ,&nbsp;Shilpy Akter ,&nbsp;Mohammad Didarul Islam ,&nbsp;MD Jubaer Bin Salim ,&nbsp;Ahamed Abdul Aziz ,&nbsp;Emranul Kabir ,&nbsp;Monir Uzzaman","doi":"10.1016/j.imu.2025.101617","DOIUrl":"10.1016/j.imu.2025.101617","url":null,"abstract":"<div><div>Naproxen (Nap), a widely used nonsteroidal anti-inflammatory drug (NSAID), effectively reduces inflammation, pain, and fever by inhibiting cyclooxygenase enzymes (i.e., COX-1 and COX-2). However, its therapeutic use is often limited by significant adverse effects, including gastrointestinal hemorrhage, nephrotoxicity, hepatotoxicity, hematuria, and aphthous ulcers. In this study, we aimed to enhance both the efficacy and safety profile of Nap by making targeted structural modifications to the parent drug. Specifically, selected functional groups (i.e., CH_3, OCH₃, CF₃, OCF₃, NH₂, CH₂NH₂, NHCONH₂ and NHCOCH₃) were introduced into the naphthalene nucleus. The geometry of the modified compounds was optimized via DFT with the B3LYP functional and 6-31+G (d, p) basis set, facilitating physicochemical and spectral analysis. Molecular docking studies were conducted against the human Prostaglandin G/H synthase 2 (5F19) and <em>Mus musculus</em> Prostaglandin-endoperoxide synthase 2 (3NT1), and these candidates were subjected to MD simulation. ADMET and PASS analyses were performed to evaluate the medicinal efficacy and toxicological profiles of the derivatives. Our findings identified several promising candidates with enhanced therapeutic benefits and reduced toxicity compared with the parent Nap. Docking analysis revealed that analogs exhibited stronger binding affinities compared to Nap and selectivity towards COX-2. These candidates demonstrated stability over a 100 ns MD simulation, exhibiting significant hydrogen bonding. Compared with the parent drug, most of these analogs displayed reduced hepatotoxicity, nephrotoxicity, carcinogenicity, and gastrointestinal hemorrhage activity, as supported by pharmacokinetic calculations. This study demonstrated that improved chemical and medicinal properties lead to a reduction in side effects.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"53 ","pages":"Article 101617"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143103458","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"End-to-end machine learning based discrimination of neoplastic and non-neoplastic intracerebral hemorrhage on computed tomography","authors":"Jawed Nawabi ,&nbsp;Sophia Schulze-Weddige ,&nbsp;Georg Lukas Baumgärtner ,&nbsp;Tobias Orth ,&nbsp;Andrea Dell'Orco ,&nbsp;Andrea Morotti ,&nbsp;Federico Mazzacane ,&nbsp;Helge Kniep ,&nbsp;Uta Hanning ,&nbsp;Michael Scheel ,&nbsp;Jens Fiehler ,&nbsp;Tobias Penzkofer","doi":"10.1016/j.imu.2025.101633","DOIUrl":"10.1016/j.imu.2025.101633","url":null,"abstract":"<div><h3>Purpose</h3><div>To develop and evaluate a fully automated segmentation and classification tool for the discrimination of neoplastic and non-neoplastic intracerebral hemorrhage (ICH) on admission Computed Tomography (CT).</div></div><div><h3>Materials and methods</h3><div>Two models were developed using a retrospective dataset of acute ICH patients with unknown etiology upon admission, based on CT scans from a single institution (January 2016 to May 2020). An nnU-Net segmentation model was trained on manually segmented ICH and perihematomal edema (PHE) masks, alongside a ResNet-34 classification model for differentiating between neoplastic and non-neoplastic ICH. The combined tool was evaluated on the test set and validated on an external cohort. Validation performance was reevaluated after enriching the training data of the segmentation model. Evaluation metrics included accuracy (Acc), area under the curve (AUC), sensitivity, specificity, and Matthews Correlation Coefficient (MCC). Performance was compared to human raters.</div></div><div><h3>Results</h3><div>Among 291 patients, 116 (39.86 %) had neoplastic and 175 (60.14 %) non-neoplastic ICH. The tool achieved an Acc of 86 % and an AUC of 85 % with a sensitivity and specificity of 80 % and 93 % in the test set. On the validation cohort (n = 58), the tool achieved an AUC of 68 % reaching 83 % after retraining of the segmentation model. The tool achieved an MCC of 0.62, compared to 0.47–0.61 for the human raters.</div></div><div><h3>Conclusion</h3><div>The tool demonstrated high diagnostic performance with potential as a decision-aiding tool; however, it relies on multi-vendor data for improved robustness, warranting further validation across diverse datasets.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"54 ","pages":"Article 101633"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143600504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Statistical analysis and machine learning of TMS-induced MEPs for predicting poststroke motor impairment and performance","authors":"Chun-Ren Phang ,&nbsp;Shintaro Uehara ,&nbsp;Sachiko Kodera ,&nbsp;Akiko Yuasa ,&nbsp;Shin Kitamura ,&nbsp;Yohei Otaka ,&nbsp;Akimasa Hirata","doi":"10.1016/j.imu.2025.101643","DOIUrl":"10.1016/j.imu.2025.101643","url":null,"abstract":"<div><div>Stroke severity is associated with the presence or absence of motor-evoked potentials (MEPs) induced by transcranial magnetic stimulation (TMS). However, there is limited evidence regarding the relationship between MEP waveforms, post-stroke motor impairment, and functional performance. This study aimed to evaluate the predictive value of inter-trial correlation (ITC), a novel metric reflecting waveform consistency, along with MEP amplitude and resting motor threshold (rMT), in estimating post-stroke motor outcomes. Thirty-eight stroke participants were enrolled, and TMS was applied to the hotspot of the first dorsal interosseous muscle in the ipsilesional or contralesional hemisphere to elicit MEPs. MEP amplitude, ITC, and rMT were analyzed in 20 participants with detectable MEPs. Pearson correlation coefficient (PCC) analysis assessed the relationships between MEP features and motor outcomes, including the Stroke Impairment Assessment Set (SIAS), Fugl-Meyer Assessment (FMA), and Action Research Arm Test (ARAT). A linear support vector machine (SVM) was trained using leave-one-subject-out cross-validation to predict the motor outcomes. Participants without detectable MEPs (n = 18) had significantly lower motor scores than those with detectable MEPs did. MEP amplitude from the contralesional side was positively correlated with SIAS, FMA, and ARAT (PCC = 0.51, 0.47, and 0.55, respectively), whereas LICI amplitude and ITC from the ipsilesional side were negatively correlated with motor scores. The SVM model predicted motor outcomes with an R² of 0.42 and a normalized root mean square error of 0.26. A Gaussian classifier achieved 75 % accuracy in classifying motor outcome improvements. These findings suggest that bilateral MEP features, particularly those from the contralesional hemisphere, offer valuable prognostic information. This study proposes a practical framework for post-stroke motor outcome prediction based on MEP analysis with potential utility in individualized rehabilitation planning.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"55 ","pages":"Article 101643"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143891006","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}

{"title":"Reducing lead requirements for wearable ECG: Chest lead reconstruction with 1D-CNN and Bi-LSTM","authors":"Kazuki Hebiguchi ,&nbsp;Hiroyoshi Togo ,&nbsp;Akimasa Hirata","doi":"10.1016/j.imu.2025.101624","DOIUrl":"10.1016/j.imu.2025.101624","url":null,"abstract":"<div><div>Wearable ECG devices encounter significant challenges in replicating the diagnostic capabilities of standard 12-lead ECGs, primarily due to the complexity of electrode placement and the need for specialized equipment. This study aims to develop a deep learning model capable of reconstructing complete 12-lead ECG waveforms using a minimal number of chest leads, thereby optimizing lead configurations for wearable ECG systems. Leveraging the PTB-XL ECG dataset, we preprocessed the signals to eliminate noise and trained a model integrating 1D convolutional layers with a Bi-directional Long Short-Term Memory (Bi-LSTM) architecture. Reconstruction performance was assessed using Pearson's correlation coefficient and root mean squared error (RMSE) across various input lead configurations, ranging from single to quintuple inputs. Our preprocessing and network architecture effectively capture both spatial and temporal features. The model achieved its highest reconstruction accuracy for leads located near the input leads, with performance gradually diminishing for more distant leads. Notably, the transitional zone between leads V₃ and V₄ presented reconstruction challenges due to polarity shifts. While increasing the number of input leads enhanced reconstruction accuracy and reduced variability, the improvements plateaued beyond the use of double input leads. Among configurations, double input leads, particularly those with two intervening leads between input pairs, offered an optimal balance between reconstruction accuracy and model complexity. This study highlights that accurate reconstruction of 12-lead ECG is achievable with only two input leads, providing a balance between diagnostic accuracy and reduced electrode requirements. These findings offer valuable insights for designing wearable ECG systems capable of reliable monitoring with fewer electrodes.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"53 ","pages":"Article 101624"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372759","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unveiling the secrets of neural network scaling for ECG classification","authors":"Byeong Tak Lee,&nbsp;Joon-myoung Kwon,&nbsp;Yong-Yeon Jo","doi":"10.1016/j.imu.2025.101639","DOIUrl":"10.1016/j.imu.2025.101639","url":null,"abstract":"<div><div>We present a new perspective on scaling neural networks for electrocardiograms (ECG). Although ResNet-based models are widely used in ECG classification, the potential benefits of network scaling remain unexplored. Our research investigates the impact of changes in the depth of layers, the number of channels, and the dimensions of the convolution kernels on performance. Contrary to computer vision practices, we found that shallower networks, with more channels and smaller kernels, lead to better performance for ECG classifications. Based on these findings, we provide insights that can guide the efficient development of models in practice. Finally, we explore why scaling hyperparameters affects ECG and computer vision differently. Our findings suggest that the inherent periodicity of the ECG signals plays a crucial role in this difference.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"55 ","pages":"Article 101639"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143839755","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}