Pub Date : 2026-03-01Epub Date: 2026-02-18DOI: 10.1016/j.compbiomed.2026.111545
Shafigh Ashrafi, Hedieh Sajedi
The StitchNet framework introduced a paradigm shift in Neural Architecture Search (NAS) by proposing the construction of neural networks from pre-trained fragments. This approach reduces computational costs and enables task-specific model creation without retraining entire networks. Building on this foundation, our study evaluates the practical application of StitchNet in constructing neural networks tailored to medical image classification tasks. Specifically, we assess its performance on a dataset of retinal images classified into three categories: healthy, dry, and wet AMD (Age-Related Macular Degeneration), namely drusen and choroidal neovascularization (CNV). By employing fragments from five pre-trained networks and integrating techniques such as recurrent neural networks (RNNs) and autoencoders, we aim to validate and enhance StitchNet's capabilities. Our findings demonstrate that while StitchNet achieves competitive accuracy with reduced computational overhead, incorporating domain-specific optimizations further improves its adaptability and efficiency. So, the developed network outperforms a scientist-designed network by 6%. In the next phase, we will explore ways to improve the algorithm's efficiency and minimize the data required for processing. Fully reproducible code here: https://github.com/ShafighAshrafi/stitchnet.
{"title":"Task-specific neural networks for medical imaging using pretrained fragments","authors":"Shafigh Ashrafi, Hedieh Sajedi","doi":"10.1016/j.compbiomed.2026.111545","DOIUrl":"10.1016/j.compbiomed.2026.111545","url":null,"abstract":"<div><div>The StitchNet framework introduced a paradigm shift in Neural Architecture Search (NAS) by proposing the construction of neural networks from pre-trained fragments. This approach reduces computational costs and enables task-specific model creation without retraining entire networks. Building on this foundation, our study evaluates the practical application of StitchNet in constructing neural networks tailored to medical image classification tasks. Specifically, we assess its performance on a dataset of retinal images classified into three categories: healthy, dry, and wet AMD (Age-Related Macular Degeneration), namely drusen and choroidal neovascularization (CNV). By employing fragments from five pre-trained networks and integrating techniques such as recurrent neural networks (RNNs) and autoencoders, we aim to validate and enhance StitchNet's capabilities. Our findings demonstrate that while StitchNet achieves competitive accuracy with reduced computational overhead, incorporating domain-specific optimizations further improves its adaptability and efficiency. So, the developed network outperforms a scientist-designed network by 6%. In the next phase, we will explore ways to improve the algorithm's efficiency and minimize the data required for processing. Fully reproducible code here: <span><span>https://github.com/ShafighAshrafi/stitchnet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111545"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146225809","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-01-30DOI: 10.1016/j.compbiomed.2026.111511
Hendrik Möller , Lukas Krautschick , Robert Graf , Matan Atad , Chia-Jung Busch , Achim Georg Beule , Christian Scharf , Lars Kaderali , Bjoern Menze , Daniel Rueckert , Jan S. Kirschke , Fabian Paperlein
Background
Chronic rhinosinusitis (CRS) is a common and persistent sinus inflammation that affects 5%–12% of the general population. It substantially reduces quality of life, yet its severity is often challenging to assess objectively. The Lund–Mackay score (LMS) rates sinus opacification but is typically assessed manually and subjectively.
Methods
We introduce Paranasal Segmentation for Imaging-based Disease Evaluation (PARASIDE), an automatic tool for segmenting air and soft tissue volumes of the structures of the sinus maxillaris, frontalis, sphenoidalis, and ethmoidalis in T1-weighted magnetic resonance imaging. Utilizing that segmentation, we quantify feature relations such as volume, thickness, and intensity relations which were previously observed only manually and subjectively. Using these features, we regress the Total Lund-Mackay Score (TLMS) of each subject. We compare our approach against established baselines: the Quantitative Opacification Score (QOS) and the Quantitative Lund–Mackay Score (QLMS).
Results
PARASIDE achieves a mean-squared error (MSE) of 2.444 and mean absolute error (MAE) of 1.181 for TLMS prediction, outperforming the QOS/QLMS baseline (MSE = 3.784, MAE = 1.445). The segmentation achieves a mean Dice similarity coefficient of 0.882 0.138 and an average symmetric surface distance (ASSD) of 0.311 0.354 mm across all structures.
Conclusion
PARASIDE enables the first automated whole-paranasal sinus segmentation for T1-weighted MRI, extracting quantitative features that predict CRS severity more accurately than existing volumetric scoring methods. By integrating high-quality segmentation with fully automated TLMS estimation, our system offers a reproducible and objective assessment tool in clinical workflows, with the potential to reduce inter-rater variability, accelerate reporting, and support large-scale retrospective studies.
慢性鼻窦炎（CRS）是一种常见的持续性鼻窦炎症，影响5%-12%的普通人群。它大大降低了生活质量，但其严重程度往往难以客观评估。隆德-麦凯评分（LMS）评价鼻窦混浊，但通常是人工和主观评估。方法引入PARASIDE (para - al Segmentation for imaging -based Disease Evaluation)，这是一种用于在t1加权磁共振成像中分割上颌窦、额肌、蝶窦和筛窦结构的空气和软组织体积的自动工具。利用这种分割，我们量化了特征关系，如体积、厚度和强度关系，这些关系以前只能手动和主观地观察到。利用这些特征，我们回归了每个受试者的总Lund-Mackay评分（TLMS）。我们将我们的方法与既定基线进行比较：定量不透明评分（QOS）和定量伦德-麦凯评分（QLMS）。结果sparaside对TLMS的预测均方误差（MSE）为2.444，平均绝对误差（MAE）为1.181，优于QOS/QLMS基线（MSE = 3.784, MAE = 1.445）。在所有结构中，分割的平均Dice相似系数为0.882±0.138，平均对称表面距离（ASSD）为0.311±0.354 mm。结论paraside首次实现了t1加权MRI的全鼻窦自动分割，提取定量特征，比现有的体积评分方法更准确地预测CRS严重程度。通过将高质量的分割与全自动TLMS评估相结合，我们的系统为临床工作流程提供了可重复和客观的评估工具，具有减少评估者之间的差异、加快报告速度和支持大规模回顾性研究的潜力。
{"title":"PARASIDE: An automatic paranasal sinus segmentation and structure analysis tool for magnetic resonance imaging","authors":"Hendrik Möller , Lukas Krautschick , Robert Graf , Matan Atad , Chia-Jung Busch , Achim Georg Beule , Christian Scharf , Lars Kaderali , Bjoern Menze , Daniel Rueckert , Jan S. Kirschke , Fabian Paperlein","doi":"10.1016/j.compbiomed.2026.111511","DOIUrl":"10.1016/j.compbiomed.2026.111511","url":null,"abstract":"<div><h3>Background</h3><div>Chronic rhinosinusitis (CRS) is a common and persistent sinus inflammation that affects 5%–12% of the general population. It substantially reduces quality of life, yet its severity is often challenging to assess objectively. The Lund–Mackay score (LMS) rates sinus opacification but is typically assessed manually and subjectively.</div></div><div><h3>Methods</h3><div>We introduce Paranasal Segmentation for Imaging-based Disease Evaluation (PARASIDE), an automatic tool for segmenting air and soft tissue volumes of the structures of the sinus maxillaris, frontalis, sphenoidalis, and ethmoidalis in T1-weighted magnetic resonance imaging. Utilizing that segmentation, we quantify feature relations such as volume, thickness, and intensity relations which were previously observed only manually and subjectively. Using these features, we regress the Total Lund-Mackay Score (TLMS) of each subject. We compare our approach against established baselines: the Quantitative Opacification Score (QOS) and the Quantitative Lund–Mackay Score (QLMS).</div></div><div><h3>Results</h3><div>PARASIDE achieves a mean-squared error (MSE) of 2.444 and mean absolute error (MAE) of 1.181 for TLMS prediction, outperforming the QOS/QLMS baseline (MSE = 3.784, MAE = 1.445). The segmentation achieves a mean Dice similarity coefficient of 0.882 <span><math><mo>±</mo></math></span> 0.138 and an average symmetric surface distance (ASSD) of 0.311 <span><math><mo>±</mo></math></span> 0.354 mm across all structures.</div></div><div><h3>Conclusion</h3><div>PARASIDE enables the first automated whole-paranasal sinus segmentation for T1-weighted MRI, extracting quantitative features that predict CRS severity more accurately than existing volumetric scoring methods. By integrating high-quality segmentation with fully automated TLMS estimation, our system offers a reproducible and objective assessment tool in clinical workflows, with the potential to reduce inter-rater variability, accelerate reporting, and support large-scale retrospective studies.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111511"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146076947","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-07DOI: 10.1016/j.compbiomed.2026.111541
Maria Suntsova , Alexander Modestov , Elizaveta Rabushko , Nikolai Komarov , Ivan Gaziev , Anastasia Novosadskaya , Anastasiya Barysionak , Galina Zakharova , Nina Shaban , Anna Khristichenko , Anna Emelianova , Marianna Zolotovskaia , Elena Poddubskaya , Alexander Seryakov , Anton Buzdin
Transposable elements (TEs) are a major source of genomic variability, yet their relation with other mutational processes and DNA repair in cancers remains poorly understood. Here we combined deep sequencing approaches (RNAseq, whole exome sequencing, and targeted TE-flank sequencing) with computational analyses to investigate the transcriptional activity of active human L1 and Alu elements across 526 experimental and 2488 TCGA cancer samples. By quantifying somatic TE insertions in 40 experimental pairs of cancer and matched normal tissues, we found that TE insertional activity (roughly 20 insertions per sample for each class of TEs) correlates with L1 transcription, is increased in cancers and has substantial intersample variability. TE insertions also correlated with activation of non-homologous end joining, mismatch and nucleotide excision repair pathways, and with transcription of TERT and APOBEC3B genes. Based on highly correlated genes, we created an expression signature reflecting TE insertional activity (AUC 0.819-0.903). On larger experimental and literature tumor cohorts, the signature strongly correlated with the activation levels of most of DNA repair pathways except those leading to ATM checkpoint activation and cell cycle arrest. It was also associated with many genome instability markers (chimeric genes, tumor mutation burden, gene copy number variation, loss of heterozygosity), but showed reduced values in cancers with microsatellite instability. Finally, the signature was associated with worse overall survival in pancreatic cancer (HR 5.9) and lesser effects in stomach, lung, and cervical cancers. These results shed light on the interplay of TE activities, DNA repair, and genome instability in human cancers.
{"title":"Insertional activity of human Alu and L1 retrotransposons is associated with DNA repair pathways and genome instability in cancer","authors":"Maria Suntsova , Alexander Modestov , Elizaveta Rabushko , Nikolai Komarov , Ivan Gaziev , Anastasia Novosadskaya , Anastasiya Barysionak , Galina Zakharova , Nina Shaban , Anna Khristichenko , Anna Emelianova , Marianna Zolotovskaia , Elena Poddubskaya , Alexander Seryakov , Anton Buzdin","doi":"10.1016/j.compbiomed.2026.111541","DOIUrl":"10.1016/j.compbiomed.2026.111541","url":null,"abstract":"<div><div>Transposable elements (TEs) are a major source of genomic variability, yet their relation with other mutational processes and DNA repair in cancers remains poorly understood. Here we combined deep sequencing approaches (RNAseq, whole exome sequencing, and targeted TE-flank sequencing) with computational analyses to investigate the transcriptional activity of active human L1 and Alu elements across 526 experimental and 2488 TCGA cancer samples. By quantifying somatic TE insertions in 40 experimental pairs of cancer and matched normal tissues, we found that TE insertional activity (roughly 20 insertions per sample for each class of TEs) correlates with L1 transcription, is increased in cancers and has substantial intersample variability. TE insertions also correlated with activation of non-homologous end joining, mismatch and nucleotide excision repair pathways, and with transcription of <em>TERT</em> and <em>APOBEC3B</em> genes. Based on highly correlated genes, we created an expression signature reflecting TE insertional activity (AUC 0.819-0.903). On larger experimental and literature tumor cohorts, the signature strongly correlated with the activation levels of most of DNA repair pathways except those leading to ATM checkpoint activation and cell cycle arrest. It was also associated with many genome instability markers (chimeric genes, tumor mutation burden, gene copy number variation, loss of heterozygosity), but showed reduced values in cancers with microsatellite instability. Finally, the signature was associated with worse overall survival in pancreatic cancer (HR 5.9) and lesser effects in stomach, lung, and cervical cancers. These results shed light on the interplay of TE activities, DNA repair, and genome instability in human cancers.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111541"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146141290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Leptospirosis is a zoonotic bacterial disease caused by Leptospira spirochetes, with limited therapeutic options, symptoms ranging from mild flu-like illness to severe organ failure and death. It presents a broad clinical spectrum, complicating diagnosis and treatment. This study targets Leptospira thiamine monophosphate kinase (ThiL), an essential enzyme conserved across the pathogenic species of Leptospira that is crucial for bacterial survival, with no known human homolog, making it a promising and selective therapeutic candidate for drug development. This study aims to discover effective inhibitors of Leptospira ThiL using Structure-Based Virtual Screening (SBVS). Potential hits were evaluated for drug-like properties, followed by Density Functional Theory (DFT) calculations to assess electronic structure properties. Further molecular dynamics simulations and binding free energy calculations were performed using the MM/PBSA approach to confirm the stability and affinity of the inhibitor. High-throughput Virtual Screening (HTVS) of phytochemicals revealed five promising candidates, namely, IMPHY004345, IMPHY005869, IMPHY006284, IMPHY002964, and IMPHY005688, exhibiting better docking scores (−12.36 to −10.54 kcal/mol) and strong MM/GBSA binding energies (−47.26 to −40.72 kcal/mol), along with optimal pharmacokinetic profiles. DFT analysis assessed the electronic properties of these compounds, providing insights into their chemical reactivity. MD simulations demonstrated stable binding and persistent hydrogen-bond interactions in the ThiL-ligand complexes. The conformational stability was monitored through MD-based distance plot analysis, revealing sustained interactions with catalytically significant residues (Glu9, Gln23, Asp39, Arg140, Thr209 Lys218) across all pathogenic Leptospira species, underscoring ThiL's evolutionary and functional importance. MM/PBSA calculations also support the high-affinity binding, with key residues emerging as crucial for maintaining complex stability and contributing to energy. This study establishes ThiL as a structurally stable, evolutionarily conserved, and highly druggable target in Leptospira. The identified leads, IMPHY006284 and IMPHY004345, emerged as the most potent broad-spectrum ThiL inhibitors, exhibiting multi-target inhibition across pathogenic species. This offers a promising strategy to overcome strain-specific variability and deliver broad-spectrum therapeutics for leptospirosis management.
{"title":"Probing the conserved catalytic mechanism of ThiL protein in pathogenic Leptospira species: An in silico strategy for inhibitor discovery to combat leptospirosis","authors":"Maheswari Narthanareeswaran , Hemavathy Nagarajan , Sneha Subramaniyan , Bhuvaneswari Narthanareeswaran , Sampathkumar Ranganathan , Jeyakanthan Jeyaraman","doi":"10.1016/j.compbiomed.2026.111540","DOIUrl":"10.1016/j.compbiomed.2026.111540","url":null,"abstract":"<div><div>Leptospirosis is a zoonotic bacterial disease caused by <em>Leptospira spirochetes</em>, with limited therapeutic options, symptoms ranging from mild flu-like illness to severe organ failure and death. It presents a broad clinical spectrum, complicating diagnosis and treatment. This study targets <em>Leptospira</em> thiamine monophosphate kinase (ThiL), an essential enzyme conserved across the pathogenic species of <em>Leptospira</em> that is crucial for bacterial survival, with no known human homolog, making it a promising and selective therapeutic candidate for drug development. This study aims to discover effective inhibitors of <em>Leptospira</em> ThiL using Structure-Based Virtual Screening (SBVS). Potential hits were evaluated for drug-like properties, followed by Density Functional Theory (DFT) calculations to assess electronic structure properties. Further molecular dynamics simulations and binding free energy calculations were performed using the MM/PBSA approach to confirm the stability and affinity of the inhibitor. High-throughput Virtual Screening (HTVS) of phytochemicals revealed five promising candidates, namely, IMPHY004345, IMPHY005869, IMPHY006284, IMPHY002964, and IMPHY005688, exhibiting better docking scores (−12.36 to −10.54 kcal/mol) and strong MM/GBSA binding energies (−47.26 to −40.72 kcal/mol), along with optimal pharmacokinetic profiles. DFT analysis assessed the electronic properties of these compounds, providing insights into their chemical reactivity. MD simulations demonstrated stable binding and persistent hydrogen-bond interactions in the ThiL-ligand complexes. The conformational stability was monitored through MD-based distance plot analysis, revealing sustained interactions with catalytically significant residues (Glu9, Gln23, Asp39, Arg140, Thr209 Lys218) across all pathogenic <em>Leptospira</em> species, underscoring ThiL's evolutionary and functional importance. MM/PBSA calculations also support the high-affinity binding, with key residues emerging as crucial for maintaining complex stability and contributing to energy. This study establishes ThiL as a structurally stable, evolutionarily conserved, and highly druggable target in <em>Leptospira</em>. The identified leads, IMPHY006284 and IMPHY004345, emerged as the most potent broad-spectrum ThiL inhibitors, exhibiting multi-target inhibition across pathogenic species. This offers a promising strategy to overcome strain-specific variability and deliver broad-spectrum therapeutics for leptospirosis management.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111540"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146149466","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Adult spinal deformity (ASD) involves complex three-dimensional (3D) spinal malalignments that impair mobility and stability. Current clinical assessments rely on static, two-dimensional (2D) radiographs, which fail to capture the 3D dynamics essential for comprehensive evaluations. While musculoskeletal models with marker-based motion analysis offer insights into kinematics, generic models fail to replicate the 3D deformities in ASD. This study introduces an automated workflow to generate image-based subject-specific models, capturing individual spinal geometry and alignment to enable analysis of 3D dynamics in patients with ASD.
Methods
A retrospective dataset of 13 deformity subjects was used to develop and evaluate the workflow. Spinopelvic bones were automatically segmented, followed by spinal joint and alignment definition. The accuracy of 3D spinal alignment was validated by simulating upright standing and bending positions as captured with biplanar radiography. 3D position and rotation differences were calculated against biplanar imaging-based reference markers.
Results
3D position differences across spinal markers averaged 2.2 ± 1.6 mm in the upright, and 3.0 ± 1.9 mm in the bending poses. In bending simulations, differences were comparable to Overbergh et al. (2020) who achieved mean errors 3.0 ± 2.0 mm. 3D rotation differences averaged 3.5 ± 1.7° in the upright, and 5.3 ± 2.6° in the bending poses. The rotation differences in bending compared well with the method of Overbergh et al. (2020) being 5.1 ± 3.0° on average.
Discussion
The proposed workflow enabled creation of image-based subject-specific models of patients with ASD, with anatomically correct spinopelvic bone geometries, intervertebral joints, and 3D alignment.
成人脊柱畸形（ASD）涉及复杂的三维（3D）脊柱错位，损害活动能力和稳定性。目前的临床评估依赖于静态的二维（2D） x线片，无法捕捉到全面评估所必需的三维动态。虽然基于标记的运动分析的肌肉骨骼模型提供了运动学的见解，但通用模型无法复制ASD的3D畸形。本研究引入了一种自动化工作流程来生成基于图像的受试者特定模型，捕获个体脊柱几何形状和对齐，从而能够分析ASD患者的3D动力学。方法：对13名残疾受试者进行回顾性数据集，以制定和评估工作流程。脊柱骨盆骨自动分割，随后是脊柱关节和对齐定义。通过模拟直立站立和弯曲位置，通过双平面x线摄影来验证3D脊柱对齐的准确性。根据基于双平面成像的参考标记计算三维位置和旋转差异。结果：脊柱标记物的三维位置差异在直立时平均为2.2±1.6 mm，在弯曲时平均为3.0±1.9 mm。在弯曲模拟中，差异与Overbergh等人（2020）相当，他们的平均误差为3.0±2.0 mm。3D旋转差异在直立时平均为3.5±1.7°，在弯曲姿势时平均为5.3±2.6°。与Overbergh et al.（2020）的方法相比，弯曲的旋转差异平均为5.1±3.0°。讨论：提出的工作流程能够创建基于图像的ASD患者特定模型，具有解剖学上正确的脊柱骨盆骨几何形状，椎间关节和3D对齐。
{"title":"Automated generation of image-based subject-specific spine models for adult spinal deformity: Development and kinematic evaluation","authors":"Birgitt Peeters , Erica Beaucage-Gauvreau , Lieven Moke , Lennart Scheys","doi":"10.1016/j.compbiomed.2026.111552","DOIUrl":"10.1016/j.compbiomed.2026.111552","url":null,"abstract":"<div><h3>Introduction</h3><div>Adult spinal deformity (ASD) involves complex three-dimensional (3D) spinal malalignments that impair mobility and stability. Current clinical assessments rely on static, two-dimensional (2D) radiographs, which fail to capture the 3D dynamics essential for comprehensive evaluations. While musculoskeletal models with marker-based motion analysis offer insights into kinematics, generic models fail to replicate the 3D deformities in ASD. This study introduces an automated workflow to generate image-based subject-specific models, capturing individual spinal geometry and alignment to enable analysis of 3D dynamics in patients with ASD.</div></div><div><h3>Methods</h3><div>A retrospective dataset of 13 deformity subjects was used to develop and evaluate the workflow. Spinopelvic bones were automatically segmented, followed by spinal joint and alignment definition. The accuracy of 3D spinal alignment was validated by simulating upright standing and bending positions as captured with biplanar radiography. 3D position and rotation differences were calculated against biplanar imaging-based reference markers.</div></div><div><h3>Results</h3><div>3D position differences across spinal markers averaged 2.2 ± 1.6 mm in the upright, and 3.0 ± 1.9 mm in the bending poses. In bending simulations, differences were comparable to Overbergh et al. (2020) who achieved mean errors 3.0 ± 2.0 mm. 3D rotation differences averaged 3.5 ± 1.7° in the upright, and 5.3 ± 2.6° in the bending poses. The rotation differences in bending compared well with the method of Overbergh et al. (2020) being 5.1 ± 3.0° on average.</div></div><div><h3>Discussion</h3><div>The proposed workflow enabled creation of image-based subject-specific models of patients with ASD, with anatomically correct spinopelvic bone geometries, intervertebral joints, and 3D alignment.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111552"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146218859","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","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.compbiomed.2026.111500
Ayan Mondal , Ayan Chatterjee , Michael A. Reigler
Gastrointestinal (GI) diseases pose significant health risks to humans. To help medical professionals in early GI disease detection and diagnosis through image processing and analysis, this article offers an in-depth exploration of improving GI disease classification through artificial intelligence, specifically focusing on convolutional neural networks (CNNs). The central objective of this research is to formulate a highly accurate model for GI disease classification. We introduce GIDNet, a novel CNN model, and present a new activation function, called Tansh, designed to improve classification accuracy. The effectiveness of the proposed approach is evaluated using the Kvasir dataset. The study addresses research gaps in the existing literature, such as limited exploration of activation functions tailored for the classification of GI diseases and lack of explainability in model decisions. The methodology section describes the experimental setup, including the implementation of the Tansh activation function, model architecture, and dataset preparation. The study conducts a comparative analysis of Tansh against well-established activation functions, evaluating classification accuracy and model explainability using well-established methods. The results reveal that the pro-posed GIDNet model integrated with the Tansh activation function achieves an unparalleled classification accuracy of 98.75 % in the Kvasir dataset, surpass-ing existing state-of-the-art models. The study concludes with discussions of the implications of the findings, potential applications in clinical practice, and avenues for future research. In general, the study contributes novel information.
on the classification of GI diseases by introducing a novel activation function and demonstrating its effectiveness in improving classification accuracy and model explainability. The findings have significant implications for automated diagno-sis and treatment planning in gastroenterology, paving the way for more reliable and interpretable AI-driven healthcare solutions.
{"title":"Gastrointestinal image classification with GIDNet CNN model and non-linear Tansh activation function","authors":"Ayan Mondal , Ayan Chatterjee , Michael A. Reigler","doi":"10.1016/j.compbiomed.2026.111500","DOIUrl":"10.1016/j.compbiomed.2026.111500","url":null,"abstract":"<div><div>Gastrointestinal (GI) diseases pose significant health risks to humans. To help medical professionals in early GI disease detection and diagnosis through image processing and analysis, this article offers an in-depth exploration of improving GI disease classification through artificial intelligence, specifically focusing on convolutional neural networks (CNNs). The central objective of this research is to formulate a highly accurate model for GI disease classification. We introduce GIDNet, a novel CNN model, and present a new activation function, called Tansh, designed to improve classification accuracy. The effectiveness of the proposed approach is evaluated using the Kvasir dataset. The study addresses research gaps in the existing literature, such as limited exploration of activation functions tailored for the classification of GI diseases and lack of explainability in model decisions. The methodology section describes the experimental setup, including the implementation of the Tansh activation function, model architecture, and dataset preparation. The study conducts a comparative analysis of Tansh against well-established activation functions, evaluating classification accuracy and model explainability using well-established methods. The results reveal that the pro-posed GIDNet model integrated with the Tansh activation function achieves an unparalleled classification accuracy of 98.75 <strong>%</strong> in the Kvasir dataset, surpass-ing existing state-of-the-art models. The study concludes with discussions of the implications of the findings, potential applications in clinical practice, and avenues for future research. In general, the study contributes novel information.</div><div>on the classification of GI diseases by introducing a novel activation function and demonstrating its effectiveness in improving classification accuracy and model explainability. The findings have significant implications for automated diagno-sis and treatment planning in gastroenterology, paving the way for more reliable and interpretable AI-driven healthcare solutions.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111500"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146137284","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-10DOI: 10.1016/j.compbiomed.2026.111536
Ana Bensabat , Marcos Gouveia , Claire Leclech , João Carvalho , Abdul I. Barakat , Rui D.M. Travasso
As they navigate complex extracellular environments, cells and their nuclei undergo extensive deformation. Recent experiments have demonstrated that vascular endothelial cells cultured on microgroove substrates, which mimic the anisotropic topography of the basement membrane, exhibit complex nuclear deformations, leading to partial or even complete nuclear penetration into the microgrooves. Interestingly, the experiments suggest that nuclear entry into the microgrooves is driven mainly by cellular adhesion and spreading rather than by cytoskeleton-mediated pulling and/or pushing forces. In the present work, we develop a phase-field model to describe endothelial cell deformation on microgroove substrates and characterize the conditions necessary for nuclear confinement within the grooves, a process that has been termed “caging" in the experiments. The model introduces a novel non-local term that prevents the cellular body from fragmenting under conditions of strong adhesion and high curvature. Our numerical simulations show that significant nuclear deformation and partial caging occur for strong cell-substrate adhesion and for nuclear membrane stiffness close to or inferior to that of the cell membrane. We further show that the dimensions of the grooves are critical for the caging process, with increasing groove depth and width favoring nuclear penetration into and caging within the grooves. These results are in close agreement with experimental observations, thus corroborating the notion that cell-substrate adhesion forces can drive large-scale nuclear deformations without the need for cytoskeleton-generated forces.
{"title":"Modeling the effect of substrate topography on cellular and nuclear deformations","authors":"Ana Bensabat , Marcos Gouveia , Claire Leclech , João Carvalho , Abdul I. Barakat , Rui D.M. Travasso","doi":"10.1016/j.compbiomed.2026.111536","DOIUrl":"10.1016/j.compbiomed.2026.111536","url":null,"abstract":"<div><div>As they navigate complex extracellular environments, cells and their nuclei undergo extensive deformation. Recent experiments have demonstrated that vascular endothelial cells cultured on microgroove substrates, which mimic the anisotropic topography of the basement membrane, exhibit complex nuclear deformations, leading to partial or even complete nuclear penetration into the microgrooves. Interestingly, the experiments suggest that nuclear entry into the microgrooves is driven mainly by cellular adhesion and spreading rather than by cytoskeleton-mediated pulling and/or pushing forces. In the present work, we develop a phase-field model to describe endothelial cell deformation on microgroove substrates and characterize the conditions necessary for nuclear confinement within the grooves, a process that has been termed “caging\" in the experiments. The model introduces a novel non-local term that prevents the cellular body from fragmenting under conditions of strong adhesion and high curvature. Our numerical simulations show that significant nuclear deformation and partial caging occur for strong cell-substrate adhesion and for nuclear membrane stiffness close to or inferior to that of the cell membrane. We further show that the dimensions of the grooves are critical for the caging process, with increasing groove depth and width favoring nuclear penetration into and caging within the grooves. These results are in close agreement with experimental observations, thus corroborating the notion that cell-substrate adhesion forces can drive large-scale nuclear deformations without the need for cytoskeleton-generated forces.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111536"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146164570","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-13DOI: 10.1016/j.compbiomed.2026.111548
Sampath Rapuri , Kirby Gong , Carl Harris , Robert D. Stevens
Background
Pulmonary embolism (PE) is a leading cause of preventable death, yet statistical prediction models have shown inconsistent validity. Our primary objective was to determine if a machine learning model trained with data routinely collected in clinical care can successfully identify acute PE in critically ill patients.
Methods
Leveraging two multicenter datasets acquired nationally (development cohort) and within the Johns Hopkins Health System (external validation cohort), we trained machine learning models with features extracted from demographics, comorbidities, physiologic and laboratory data available following intensive care unit (ICU) admission. The primary endpoint was the identification of acute PE during ICU admission. Model performance was contrasted with two benchmark PE risk scores.
Findings
PE was diagnosed in 2647 of 164,383 (1.61%) and 754 of 64,923 admissions (1.16%) in the development and external validation datasets respectively. Using data from the first 48 h after ICU admission, the mean (95% CI) discrimination measured by area under the receiver characteristic curve (AUROC) was 0.829 (0.808–0.852), 0.704 (0.681–0.727), and 0.667 (0.653–0.681) for our logistic regression machine learning model and for the two benchmark scores, respectively; mean area under the precision recall curve was 0.150 (0.138–0.162), 0.080 (0.071–0.089), and 0.081 (0.071–0.091), respectively. Discrimination was maintained in the external validation dataset with an AUROC of 0.819 (0.802–0.836).
Interpretation
Findings indicate that PE can be detected accurately in ICU patients using routinely collected clinical data. The machine learning model successfully validated and outperformed existing benchmark risk scores. Such a model could become a valuable tool for assessing the likelihood of PE among critically ill patients.
{"title":"A machine learning model to identify pulmonary embolism in patients admitted to intensive care","authors":"Sampath Rapuri , Kirby Gong , Carl Harris , Robert D. Stevens","doi":"10.1016/j.compbiomed.2026.111548","DOIUrl":"10.1016/j.compbiomed.2026.111548","url":null,"abstract":"<div><h3>Background</h3><div>Pulmonary embolism (PE) is a leading cause of preventable death, yet statistical prediction models have shown inconsistent validity. Our primary objective was to determine if a machine learning model trained with data routinely collected in clinical care can successfully identify acute PE in critically ill patients.</div></div><div><h3>Methods</h3><div>Leveraging two multicenter datasets acquired nationally (development cohort) and within the Johns Hopkins Health System (external validation cohort), we trained machine learning models with features extracted from demographics, comorbidities, physiologic and laboratory data available following intensive care unit (ICU) admission. The primary endpoint was the identification of acute PE during ICU admission. Model performance was contrasted with two benchmark PE risk scores.</div></div><div><h3>Findings</h3><div>PE was diagnosed in 2647 of 164,383 (1.61%) and 754 of 64,923 admissions (1.16%) in the development and external validation datasets respectively. Using data from the first 48 h after ICU admission, the mean (95% CI) discrimination measured by area under the receiver characteristic curve (AUROC) was 0.829 (0.808–0.852), 0.704 (0.681–0.727), and 0.667 (0.653–0.681) for our logistic regression machine learning model and for the two benchmark scores, respectively; mean area under the precision recall curve was 0.150 (0.138–0.162), 0.080 (0.071–0.089), and 0.081 (0.071–0.091), respectively. Discrimination was maintained in the external validation dataset with an AUROC of 0.819 (0.802–0.836).</div></div><div><h3>Interpretation</h3><div>Findings indicate that PE can be detected accurately in ICU patients using routinely collected clinical data. The machine learning model successfully validated and outperformed existing benchmark risk scores. Such a model could become a valuable tool for assessing the likelihood of PE among critically ill patients.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111548"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146171746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-03-01Epub Date: 2026-02-09DOI: 10.1016/j.compbiomed.2026.111538
Anna Josefine Grillenberger , Nelly Shenton , Martin Lauritzen , Krisztina Benedek , Sadasivan Puthusserypady
Objective
Detecting Alzheimer’s disease (AD) at an early stage is essential for administering effective treatments and preventing neuronal damage. Unfortunately, current diagnostic techniques are often invasive and expensive. Our research focuses on creating a cost-effective and non-invasive method for the early detection of cognitive decline.
Methods
Using a publicly available dataset of resting state electroencephalographic (EEG) data on healthy controls and patients with Mild Cognitive Impairment (MCI), two novel deep learning (DL) algorithms with self-attention mechanisms were developed and evaluated for their performance in predicting MCI and cognitive decline.
Results
Both proposed DL algorithms outperformed a traditional convolutional neural network (CNN) model in predicting MCI, achieving test accuracy improvements of 8.5% and 10%, respectively, while utilizing significantly fewer trainable parameters. An ablation study highlighted the attention layer as a key feature, enhancing model accuracy by 8.5%. Analysis of the attention layers indicated that beta band frequencies (13-30 Hz) were essential for distinguishing MCI from control subjects, highlighting the role of high EEG frequencies in early cognitive deficits. Predicting pre-clinical cognitive decline in healthy subjects proved more challenging than predicting diagnosed MCI. However, using transfer-learning methods, we achieved a test accuracy of 56.08%.
Conclusion
Our models achieved state-of-the-art results in the MCI classification task, and demonstrated learning progress in predicting cognitive decline in the preclinical stage. As this is the first time DL models have been evaluated to classify healthy subjects based on cognitive scores, where brain changes are minimal and difficult to detect, this study opens new avenues for discovering biomarkers in early AD diagnosis and facilitating early interventions. Interpretation of the trained DL attention models provided valuable insights that aligned with the existing brain research, serving as a helpful tool for validating AI in healthcare applications.
{"title":"Exploring the potential of explainable deep learning for EEG-based cognitive decline prediction","authors":"Anna Josefine Grillenberger , Nelly Shenton , Martin Lauritzen , Krisztina Benedek , Sadasivan Puthusserypady","doi":"10.1016/j.compbiomed.2026.111538","DOIUrl":"10.1016/j.compbiomed.2026.111538","url":null,"abstract":"<div><h3>Objective</h3><div>Detecting Alzheimer’s disease (AD) at an early stage is essential for administering effective treatments and preventing neuronal damage. Unfortunately, current diagnostic techniques are often invasive and expensive. Our research focuses on creating a cost-effective and non-invasive method for the early detection of cognitive decline.</div></div><div><h3>Methods</h3><div>Using a publicly available dataset of resting state electroencephalographic (EEG) data on healthy controls and patients with Mild Cognitive Impairment (MCI), two novel deep learning (DL) algorithms with self-attention mechanisms were developed and evaluated for their performance in predicting MCI and cognitive decline.</div></div><div><h3>Results</h3><div>Both proposed DL algorithms outperformed a traditional convolutional neural network (CNN) model in predicting MCI, achieving test accuracy improvements of 8.5% and 10%, respectively, while utilizing significantly fewer trainable parameters. An ablation study highlighted the attention layer as a key feature, enhancing model accuracy by 8.5%. Analysis of the attention layers indicated that beta band frequencies (13-30 Hz) were essential for distinguishing MCI from control subjects, highlighting the role of high EEG frequencies in early cognitive deficits. Predicting pre-clinical cognitive decline in healthy subjects proved more challenging than predicting diagnosed MCI. However, using transfer-learning methods, we achieved a test accuracy of 56.08%.</div></div><div><h3>Conclusion</h3><div>Our models achieved state-of-the-art results in the MCI classification task, and demonstrated learning progress in predicting cognitive decline in the preclinical stage. As this is the first time DL models have been evaluated to classify healthy subjects based on cognitive scores, where brain changes are minimal and difficult to detect, this study opens new avenues for discovering biomarkers in early AD diagnosis and facilitating early interventions. Interpretation of the trained DL attention models provided valuable insights that aligned with the existing brain research, serving as a helpful tool for validating AI in healthcare applications.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"204 ","pages":"Article 111538"},"PeriodicalIF":6.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146156339","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","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