Food Chem X. 2024 Nov 13;24:101981. doi: 10.1016/j.fochx.2024.101981. eCollection 2024 Dec 30.

ABSTRACT

With the rising demand of saffron, it is essential to standardize the confirmation of its origin and identify any adulteration to maintain a good quality led market product. However, a rapid and reliable strategy for identifying the adulteration saffron is still lacks. Herein, a combination of headspace-gas chromatography-ion mobility spectrometry (HS-GC-IMS) and convolutional neural network (CNN) was developed. Sixty-nine volatile compounds (VOCs) including 7 groups of isomers were detected rapidly and directly. A CNN prediction model based on GC-IMS data was proposed. With the merit of minimal data prepossessing and automatic feature extraction capability, GC-IMS images were directly input to the CNN model. The origin prediction results were output with the average accuracy about 90 %, which was higher than traditional methods like PCA (61 %) and SVM (71 %). This established CNN also showed ability in identifying counterfeit saffron with a high accuracy of 98 %, which can be used to authenticate saffron.

PMID:39850938 | PMC:PMC11754009 | DOI:10.1016/j.fochx.2024.101981

Front Public Health. 2025 Jan 15;12:1426168. doi: 10.3389/fpubh.2024.1426168. eCollection 2024.

ABSTRACT

INTRODUCTION: The growing demand for real-time, affordable, and accessible healthcare has underscored the need for advanced technologies that can provide timely health monitoring. One such area is predicting arterial blood pressure (BP) using non-invasive methods, which is crucial for managing cardiovascular diseases. This research aims to address the limitations of current healthcare systems, particularly in remote areas, by leveraging deep learning techniques in Smart Health Monitoring (SHM).

METHODS: This paper introduces a novel neural network architecture, ResNet-LSTM, to predict BP from physiological signals such as electrocardiogram (ECG) and photoplethysmogram (PPG). The combination of ResNet's feature extraction capabilities and LSTM's sequential data processing offers improved prediction accuracy. Comprehensive error analysis was conducted, and the model was validated using Leave-One-Out (LOO) cross-validation and an additional dataset.

RESULTS: The ResNet-LSTM model showed superior performance, particularly with PPG data, achieving a mean absolute error (MAE) of 6.2 mmHg and a root mean square error (RMSE) of 8.9 mmHg for BP prediction. Despite the higher computational cost (~4,375 FLOPs), the improved accuracy and generalization across datasets demonstrate the model's robustness and suitability for continuous BP monitoring.

DISCUSSION: The results confirm the potential of integrating ResNet-LSTM into SHM for accurate and non-invasive BP prediction. This approach also highlights the need for accurate anomaly detection in continuous monitoring systems, especially for wearable devices. Future work will focus on enhancing cloud-based infrastructures for real-time analysis and refining anomaly detection models to improve patient outcomes.

PMID:39850864 | PMC:PMC11755415 | DOI:10.3389/fpubh.2024.1426168

Front Artif Intell. 2025 Jan 9;7:1498956. doi: 10.3389/frai.2024.1498956. eCollection 2024.

ABSTRACT

INTRODUCTION: Active learning can significantly decrease the labeling cost of deep learning workflows by prioritizing the limited labeling budget to high-impact data points that have the highest positive impact on model accuracy. Active learning is especially useful for semantic segmentation tasks where we can selectively label only a few high-impact regions within these high-impact images. Most established regional active learning algorithms deploy a static-budget querying strategy where a fixed percentage of regions are queried in each image. A static budget could result in over- or under-labeling images as the number of high-impact regions in each image can vary.

METHODS: In this paper, we present a novel dynamic-budget superpixel querying strategy that can query the optimal numbers of high-uncertainty superpixels in an image to improve the querying efficiency of regional active learning algorithms designed for semantic segmentation.

RESULTS: For two distinct datasets, we show that by allowing a dynamic budget for each image, the active learning algorithm is more effective compared to static-budget querying at the same low total labeling budget. We investigate both low- and high-budget scenarios and the impact of superpixel size on our dynamic active learning scheme. In a low-budget scenario, our dynamic-budget querying outperforms static-budget querying by 5.6% mIoU on a specialized agriculture field image dataset and 2.4% mIoU on Cityscapes.

DISCUSSION: The presented dynamic-budget querying strategy is simple, effective, and can be easily adapted to other regional active learning algorithms to further improve the data efficiency of semantic segmentation tasks.

PMID:39850848 | PMC:PMC11754207 | DOI:10.3389/frai.2024.1498956

Int J Burns Trauma. 2024 Dec 15;14(6):125-132. doi: 10.62347/VCJP9652. eCollection 2024.

ABSTRACT

PURPOSE: To evaluate the identification of nasal bone fractures and their clinical diagnostic significance for three-dimensional (3D) reconstruction of maxillofacial computed tomography (CT) images by applying artificial intelligence (AI) with deep learning (DL).

METHODS: CT maxillofacial 3D reconstruction images of 39 patients with normal nasal bone and 43 patients with nasal bone fracture were retrospectively analysed, and a total of 247 images were obtained in three directions: the orthostatic, left lateral and right lateral positions. The CT scan images of all patients were reviewed by two senior specialists to confirm the presence or absence of nasal fractures. Binary classification prediction was performed using the YOLOX detection model + GhostNetv2 classification model with a DL algorithm. Accuracy, sensitivity, and specificity were used to evaluate the efficacy of the AI model. Manual independent review, and AI model-assisted manual independent review were used to identify nasal fractures.

RESULTS: Compared with those of manual independent detection, the accuracy, sensitivity, and specificity of AI-assisted film reading improved between junior and senior physicians. The differences were statistically significant (P<0.05), and all were higher than manual independent detection.

CONCLUSIONS: Based on deep learning methods, an artificial intelligence model can be used to assist in the diagnosis of nasal bone fractures, which helps to promote the practical clinical application of deep learning methods.

PMID:39850782 | PMC:PMC11751554 | DOI:10.62347/VCJP9652

Int J Cardiol Heart Vasc. 2025 Jan 2;56:101593. doi: 10.1016/j.ijcha.2024.101593. eCollection 2025 Feb.

ABSTRACT

BACKGROUND: The National Lung Screening Trial (NLST) has shown that screening with low dose CT in high-risk population was associated with reduction in lung cancer mortality. These patients are also at high risk of coronary artery disease, and we used deep learning model to automatically detect, quantify and perform risk categorisation of coronary artery calcification score (CACS) from non-ECG gated Chest CT scans.

MATERIALS AND METHODS: Automated calcium quantification was performed using a neural network based on Mask regions with convolutional neural networks (R-CNN) for multiorgan segmentation. Manual evaluation of calcium was carried out using proprietary software. This study used 80 patients to train the segmentation model and randomly selected 1442 patients were used for the validation of the algorithm. We compared the model generated results with Ground Truth.

RESULTS: Automatic cardiac and aortic segmentation model worked well (Mean Dice score: 0.91). Cohen's kappa coefficient between the reference actual and the interclass computed predictive categories on the test set is 0.72 (95 % CI: 0.61-0.83). Our method correctly classifies the risk group in 78.8 % of the cases and classifies the subjects in the same group. F-score is measured as 0.78; 0.71; 0.81; 0.82; 0.92 in calcium score categories 0(CS:0), I (1-99), II (100-400), III (400-1000), IV (>1000), respectively. 79 % of the predictive scores lie in the same categories, 20 % of the predictive scores are one category up or down, and only 1.2 % patients were more than one category off. For the presence/absence of coronary artery calcifications, our deep learning model achieved a sensitivity of 90 % and a specificity of 94 %.

CONCLUSION: Fully automated model shows good correlation compared with reference standards. Automating the process could improve diagnostic ability, risk categorization, facilitate primary prevention intervention, improve morbidity and mortality, and decrease healthcare costs.

PMID:39850777 | PMC:PMC11754490 | DOI:10.1016/j.ijcha.2024.101593

Proc (IEEE Conf Multimed Inf Process Retr). 2024 Aug;2024:369-375. doi: 10.1109/mipr62202.2024.00065. Epub 2024 Oct 15.

ABSTRACT

Despite the prevalence of images and texts in machine learning, tabular data remains widely used across various domains. Existing deep learning models, such as convolutional neural networks and transformers, perform well however demand extensive preprocessing and tuning limiting accessibility and scalability. This work introduces an innovative approach based on a structured state-space model (SSM), MambaTab, for tabular data. SSMs have strong capabilities for efficiently extracting effective representations from data with long-range dependencies. MambaTab leverages Mamba, an emerging SSM variant, for end-to-end supervised learning on tables. Compared to state-of-the-art baselines, MambaTab delivers superior performance while requiring significantly fewer parameters, as empirically validated on diverse benchmark datasets. MambaTab's efficiency, scalability, generalizability, and predictive gains signify it as a lightweight, "plug-and-play" solution for diverse tabular data with promise for enabling wider practical applications.

PMID:39850741 | PMC:PMC11755428 | DOI:10.1109/mipr62202.2024.00065

Int J Remote Sens. 2024 Nov 19;46(2):811-834. doi: 10.1080/01431161.2024.2423909. eCollection 2025.

ABSTRACT

Tidal flat ecosystems, are under steady decline due to anthropogenic pressures including sea level rise and climate change. Monitoring and managing these coastal systems requires accurate and up-to-date mapping. Sediment characteristics and macrozoobenthos are major indicators of the environmental status of tidal flats. Field monitoring of these indicators is often restricted by low accessibility and high costs. Despite limitations in spectral contrast, integrating remote sensing with deep learning proved efficient for deriving macrozoobenthos and sediment properties. In this study, we combined deep-learning features derived from Sentinel-2 images and Object-Based Image Analysis (OBIA) to explicitly include spatial aspects in the prediction of tsediment and macrozoobenthos properties of tidal flats , as well as the distribution of four benthic species. The deep-learning features extracted from a convolutional autoencoder model were analysed with OBIA to include spatial, textural, and contextual information. Object sets of varying sizes and shapes based on the spectral bands and/or the deep-learning features, served as the spatial units. These object sets and the field-collected points were used to train the Random Forest prediction model. Predictions were made for the tidal basins Pinkegat and Zoutkamperlaag in the Dutch Wadden Sea for 2018 to 2020. The overall prediction scores of the environmental variables ranged between 0.31 and 0.54. The species-distribution prediction model achieved accuracies ranging from 0.54 to 0.68 for the four benthic species). There was an average improvement of 21% points on predictions using objects with deep learning features compared to the pixel-based predictions with just the spectral bands. The mean spatial unit that captured the patterns best ranged between 0.3 ha and 13 ha for the different variables. Overall, using both OBIA and deep-learning features consistently improved the predictions, making it a valuable combination for monitoring these important environmental variables of coastal regions.

PMID:39850715 | PMC:PMC11755323 | DOI:10.1080/01431161.2024.2423909

Indian Dermatol Online J. 2024 Dec 11;16(1):40-49. doi: 10.4103/idoj.idoj_460_24. eCollection 2025 Jan-Feb.

ABSTRACT

BACKGROUND: Artificial intelligence (AI) is revolutionizing healthcare by enabling systems to perform tasks traditionally requiring human intelligence. In healthcare, AI encompasses various subfields, including machine learning, deep learning, natural language processing, and expert systems. In the specific domain of onychology, AI presents a promising avenue for diagnosing nail disorders, analyzing intricate patterns, and improving diagnostic accuracy. This review provides a comprehensive overview of the current applications of AI in onychology, focusing on its role in diagnosing onychomycosis, subungual melanoma, nail psoriasis, nail fold capillaroscopy, and nail involvement in systemic diseases.

MATERIALS AND METHODS: A literature review on AI in nail disorders was conducted via PubMed and Google Scholar, yielding relevant studies. AI algorithms, particularly deep convolutional neural networks (CNNs), have demonstrated high sensitivity and specificity in interpreting nail images, aiding differential diagnosis as well as enhancing the efficiency of diagnostic processes in a busy clinical setting. In studies evaluating onychomycosis, AI has shown the ability to distinguish between normal nails, fungal infections, and other differentials, including nail psoriasis, with a high accuracy. AI systems have proven effective in identifying subungual melanoma. For nail psoriasis, AI has been used to automate the scoring of disease severity, reducing the time and effort required. AI applications in nail fold capillaroscopy have aided the analysis of diagnosis and prognosis of connective tissue diseases. AI applications have also been extended to recognize nail manifestations of systemic diseases, by analyzing changes in nail morphology and coloration. AI also facilitates the management of nail disorders by offering tools for personalized treatment planning, remote care, treatment monitoring, and patient education.

CONCLUSION: Despite these advancements, challenges such as data scarcity, image heterogeneity, interpretability issues, regulatory compliance, and poor workflow integration hinder the seamless adoption of AI in onychology practice. Ongoing research and collaboration between AI developers and nail experts is crucial to realize the full potential of AI in improving patient outcomes in onychology.

PMID:39850679 | PMC:PMC11753549 | DOI:10.4103/idoj.idoj_460_24

Comput Struct Biotechnol J. 2024 Dec 30;27:160-170. doi: 10.1016/j.csbj.2024.12.015. eCollection 2025.

ABSTRACT

Quality estimation of the predicted interaction interface of protein complex structural models is not only important for complex model evaluation and selection but also useful for protein-protein docking. Despite recent progress fueled by symmetry-aware deep learning architectures and pretrained protein language models (pLMs), existing methods for estimating protein complex quality have yet to fully exploit the collective potentials of these advances for accurate estimation of protein-protein interface. Here we present EquiRank, an improved protein-protein interface quality estimation method by leveraging the strength of a symmetry-aware E(3) equivariant deep graph neural network (EGNN) and integrating pLM embeddings from the pretrained ESM-2 model. Our method estimates the quality of the protein-protein interface through an effective graph-based representation of interacting residue pairs, incorporating a diverse set of features, including ESM-2 embeddings, and then by learning the representation using symmetry-aware EGNNs. Our experimental results demonstrate improved ranking performance on diverse datasets over existing latest protein complex quality estimation methods including the top-performing CASP15 protein complex quality estimation method VoroIF_GNN and the self-assessment module of AlphaFold-Multimer repurposed for protein complex scoring and across different performance evaluation metrics. Additionally, our ablation studies demonstrate the contributions of both pLMs and the equivariant nature of EGNN for improved protein-protein interface quality estimation performance. EquiRank is freely available at https://github.com/mhshuvo1/EquiRank.

PMID:39850657 | PMC:PMC11755013 | DOI:10.1016/j.csbj.2024.12.015

Acad Radiol. 2025 Jan 22:S1076-6332(24)00968-1. doi: 10.1016/j.acra.2024.12.020. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: Preoperative assessment of axillary lymph node (ALN) status is essential for breast cancer management. This study explores the use of photoacoustic (PA) imaging combined with attention-guided deep learning (DL) for precise prediction of ALN status.

MATERIALS AND METHODS: This retrospective study included patients with histologically confirmed early-stage breast cancer from 2022 to 2024, randomly divided (8:2) into training and test cohorts. All patients underwent preoperative dual modal photoacoustic-ultrasound (PA-US) examination, were treated with surgery and sentinel lymph node biopsy or ALN dissection, and were pathologically examined to determine the ALN status. Attention-guided DL model was developed using PA-US images to predict ALN status. A clinical model, constructed via multivariate logistic regression, served as the baseline for comparison. Subsequently, a nomogram incorporating the DL model and independent clinical parameters was developed. The performance of the models was evaluated through discrimination, calibration, and clinical applicability.

RESULTS: A total of 324 patients (mean age ± standard deviation, 51.0 ± 10.9 years) were included in the study and were divided into a development cohort (n = 259 [79.9%]) and a test cohort (n = 65 [20.1%]). The clinical model incorporating three independent clinical parameters yielded an area under the curve (AUC) of 0.775 (95% confidence interval [CI], 0.711-0.829) in the training cohort and 0.783 (95% CI, 0.654-0.897) in the test cohort for predicting ALN status. In comparison, the nomogram showed superior predictive performance, with an AUC of 0.906 (95% CI, 0.867-0.940) in the training cohort and 0.868 (95% CI, 0.769-0.954) in the test cohort. Decision curve analysis further confirmed the nomogram's clinical applicability, demonstrating a better net benefit across relevant threshold probabilities.

CONCLUSION: This study highlights the effectiveness of attention-guided PA imaging in breast cancer patients, providing novel nomograms for individualized clinical decision-making in predicting ALN node status.

PMID:39848886 | DOI:10.1016/j.acra.2024.12.020

AJNR Am J Neuroradiol. 2025 Jan 23. doi: 10.3174/ajnr.A8493. Online ahead of print.

ABSTRACT

BACKGROUND AND PURPOSE: Whether differences in the O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation status of glioblastoma (GBM) are reflected in MRI markers remains largely unknown. In this work, we analyze the ADC in the perienhancing infiltration zone of GBM according to the corresponding MGMT status by using a novel distance-resolved 3D evaluation.

MATERIALS AND METHODS: One hundred one patients with IDH wild-type GBM were retrospectively analyzed. GBM was segmented in 3D with deep learning. Tissue with FLAIR hyperintensity around the contrast-enhanced tumor was divided into concentric distance-resolved subvolumes. Mean ADC was calculated for the 3D tumor core and for the distance-resolved volumes around the core. Differences in group mean ADC between patients with MGMT promoter methylated (mMGMT, n = 43) and MGMT promoter unmethylated (uMGMT, n = 58) GBM was analyzed with Wilcoxon signed rank test.

RESULTS: For both mMGMT and uMGMT GBM, mean ADC values around the tumor core significantly increased as a function of distance from the core toward the periphery of the perienhancing FLAIR hyperintensity (approximately 10% increase within 5 voxels with P < 001). While group mean ADC in the tumor core was not significantly different, the distance-resolved ADC profile around the core was approximately 10% higher in mMGMT than in uMGMT GBM (P < 10-8 at 5 voxel distance from the tumor core).

CONCLUSIONS: Distance-resolved volumetric ADC analysis around the tumor core reveals tissue signatures of GBM imperceptible to the human eye on conventional MRI. The different ADC profiles around the core suggest epigenetically influenced differences in perienhancing tissue characteristics between mMGMT and uMGMT GBM.

PMID:39848779 | DOI:10.3174/ajnr.A8493

J Hepatol. 2025 Jan 21:S0168-8278(25)00018-2. doi: 10.1016/j.jhep.2025.01.011. Online ahead of print.

ABSTRACT

BACKGROUND & AIMS: Accurate multi-classification is the prerequisite for reasonable management of focal liver lesions (FLLs). Ultrasound is the common image examination, but lacks accuracy. Contrast enhanced ultrasound (CEUS) offers better performance, but highly relies on experience. Therefore, we aimed to develop a CEUS-based artificial intelligence (AI) model for FLL multi-classification and evaluate its performance in multicenter clinical tests.

METHODS: Since January 2017 to December 2023, CEUS videos, immunohistochemical biomarkers and clinical information of solid FLLs>1cm in adults were collected from 52 centers to build and test the model. It aimed to classify FLLs into six types: hepatocellular carcinoma, hepatic metastasis, intrahepatic cholangiocarcinoma, hepatic hemangioma, hepatic abscess and others. First, Module-Disease, Module-Biomarker and Module-Clinic were built in training set A and validation set. Then, three modules were aggregated as Model-DCB in training set B and internal test set. Model-DCB performance was compared with CEUS and MRI radiologists in three external test sets.

RESULTS: In total 3725 FLLs from 52 centers were divided into training set A (n=2088), validation set (n=592), training set B (n=234), internal test set (n=110), external test set A (n=113), B (n=276) and C (n=312). In external test sets A, B and C, Model-DCB all achieved significantly better performance (Accuracy from 0.85 to 0.86) than junior CEUS-radiologists (0.59-0.73), and comparable to senior CEUS-radiologists (0.79-0.85) and senior MRI-radiologists (0.82-0.86). In multiple subgroup analyses on demographic characteristics, tumor characteristics and ultrasound devices, its accuracy ranged from 0.79 to 0.92.

CONCLUSIONS: CEUS-based Model-DCB provides accurate multi-classification of FLLs. It holds promise to benefit a wide range of population, especially for patients in remote suburban areas who have difficulty accessing MRI.

IMPACT AND IMPLICATIONS: Ultrasound is the most common image examination for screening focal liver lesions (FLLs), but it lacks accuracy for multi-classification, which is the prerequisite for reasonable management. Contrast enhanced ultrasound (CEUS) offers better diagnostic performance, but highly relies on work experience of radiologists. We develop a CEUS-based model (Model-DCB) can assist junior CEUS radiologists to achieve comparable diagnostic ability to senior CEUS radiologists and senior MRI radiologists. The combination of ultrasound device, CEUS examination and Model-DCB enables even patients in remote areas to obtain excellent diagnostic performance through examination by junior radiologists.

CLINICAL TRIAL: NCT04682886.

PMID:39848548 | DOI:10.1016/j.jhep.2025.01.011

Drug Alcohol Depend. 2025 Jan 15;268:112554. doi: 10.1016/j.drugalcdep.2025.112554. Online ahead of print.

ABSTRACT

BACKGROUND: The hypothalamus is involved in stress regulation and reward processing, with its various nuclei exhibiting unique functions and connections. However, human neuroimaging studies on the hypothalamic subregions are limited in drug addiction. This study examined the volumes and functional connectivity of hypothalamic subregions in individuals with alcohol use disorder (AUD).

METHOD: The study included 24 male patients with AUD who had maintained abstinence and 24 healthy male controls, all of whom underwent brain structural and resting-state functional magnetic resonance imaging. The hypothalamus was segmented into five subunits using a deep learning-based algorithm, with comparisons of volumes and functional connectivity (FC) between the two groups. The relationships between these measures and alcohol-related characteristics were examined in the AUD group.

RESULTS: Findings indicated lower volumes in the anterior-superior (corrected-p < 0.001) and tuberal-superior subunits (corrected-p = 0.002) and altered FC of these and the anterior-inferior subunit among AUD patients (corrected-p < 0.05). Moreover, greater disease severity and a longer history of heavy drinking correlated with lower volumes in the anterior-superior (r = -0.42, p = 0.045) and tuberal-superior subregions (r = -0.61, p = 0.013), respectively. Conversely, a longer abstinence duration was associated with larger volumes in the anterior-superior (r = 0.56, p = 0.008) and tuberal-superior subunits (r = 0.40, p = 0.048) and with higher FC between the tuberal-superior hypothalamus and the thalamus, caudate, and anterior cingulate cortex (r = 0.55, p = 0.014).

CONCLUSIONS: Our results suggest that specific regional alterations within the hypothalamus, particularly the superior subregions, are associated with AUD, and more importantly, that these alterations may be reversible with prolonged abstinence.

PMID:39848134 | DOI:10.1016/j.drugalcdep.2025.112554

Comput Methods Programs Biomed. 2025 Jan 20;261:108606. doi: 10.1016/j.cmpb.2025.108606. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Atrial fibrillation (AF) is a significant cause of life-threatening heart disease due to its potential to lead to stroke and heart failure. Although deep learning-assisted diagnosis of AF based on ECG holds significance in clinical settings, it remains unsatisfactory due to insufficient consideration of noise and redundant features. In this work, we propose a novel multiscale feature-enhanced gating network (MFEG Net) for AF diagnosis.

METHOD: The network integrates multiscale convolution, adaptive feature enhancement (FE), and dynamic temporal processing. The multiscale convolution helps capture global and local information. The FE module consists of a soft-threshold residual shrinkage component, a dilated convolution module, and a Squeeze-and-Excitation (SE) module, eliminating redundant features and emphasizing effective features. The design allows the network to focus on the most relevant AF features, thereby enhancing its robustness and accuracy in the presence of noise and irrelevant information. The dynamic temporal module helps the network learn and recognize the time dependence associated with AF. The novel design endows the model with excellent robustness to cope with random noise in real-world environments.

RESULT: Compared with the state-of-the-art methods, our model exhibits excellent classification performance with an accuracy of 0.930, an F1 score of 0.883, and remarkable resilience to noise interference on the PhysioNet Challenge 2017 dataset. Moreover, the model was trained on the CinC2017 database and validated on the CPSC2018 database and AFDB database, achieving accuracies of 0.908 and 0.938, respectively.

CONCLUSION: The excellent classification performance of MFEG Net, coupled with its robustness in processing noisy electrocardiogram signals, makes it a powerful method for automatic atrial fibrillation detection. This method has made significant progress over state-of-the-art methods and may alleviate the burden of manual diagnosis for clinical doctors.

PMID:39847993 | DOI:10.1016/j.cmpb.2025.108606

Int J Comput Assist Radiol Surg. 2025 Jan 23. doi: 10.1007/s11548-024-03315-8. Online ahead of print.

ABSTRACT

PURPOSE: Semantic segmentation and landmark detection are fundamental tasks of medical image processing, facilitating further analysis of anatomical objects. Although deep learning-based pixel-wise classification has set a new-state-of-the-art for segmentation, it falls short in landmark detection, a strength of shape-based approaches.

METHODS: In this work, we propose a dense image-to-shape representation that enables the joint learning of landmarks and semantic segmentation by employing a fully convolutional architecture. Our method intuitively allows the extraction of arbitrary landmarks due to its representation of anatomical correspondences. We benchmark our method against the state-of-the-art for semantic segmentation (nnUNet), a shape-based approach employing geometric deep learning and a convolutional neural network-based method for landmark detection.

RESULTS: We evaluate our method on two medical datasets: one common benchmark featuring the lungs, heart, and clavicle from thorax X-rays, and another with 17 different bones in the paediatric wrist. While our method is on par with the landmark detection baseline in the thorax setting (error in mm of 2.6 ± 0.9 vs. 2.7 ± 0.9 ), it substantially surpassed it in the more complex wrist setting ( 1.1 ± 0.6 vs. 1.9 ± 0.5 ).

CONCLUSION: We demonstrate that dense geometric shape representation is beneficial for challenging landmark detection tasks and outperforms previous state-of-the-art using heatmap regression. While it does not require explicit training on the landmarks themselves, allowing for the addition of new landmarks without necessitating retraining.

PMID:39849288 | DOI:10.1007/s11548-024-03315-8

Neurosurg Rev. 2025 Jan 24;48(1):78. doi: 10.1007/s10143-025-03236-3.

ABSTRACT

Currently, the World Health Organization (WHO) grade of meningiomas is determined based on the biopsy results. Therefore, accurate non-invasive preoperative grading could significantly improve treatment planning and patient outcomes. Considering recent advances in machine learning (ML) and deep learning (DL), this meta-analysis aimed to evaluate the performance of these models in predicting the WHO meningioma grade using imaging data. A systematic search was performed in PubMed/MEDLINE, Embase, and the Cochrane Library for studies published up to April 1, 2024, and reporting the performance metrics of the ML models in predicting of WHO meningioma grade using imaging studies. Pooled area under the receiver operating characteristics curve (AUROC), specificity, and sensitivity were estimated. Subgroup and meta-regression analyses were performed based on a number of potential influencing variables. A total of 32 studies with 15,365 patients were included in the present study. The overall pooled sensitivity, specificity, and AUROC of ML methods for prediction of tumor grade in meningioma were 85% (95% CI, 79-89%), 87% (95% CI, 81-91%), and 93% (95% CI, 90-95%), respectively. Both the type of validation and study cohort (training or test) were significantly associated with model performance. However, no significant association was found between the sample size or the type of ML method and model performance. The ML predictive models show a high overall performance in predicting the WHO meningioma grade using imaging data. Further studies on the performance of DL algorithms in larger datasets using external validation are needed.

PMID:39849257 | DOI:10.1007/s10143-025-03236-3

Med Biol Eng Comput. 2025 Jan 24. doi: 10.1007/s11517-024-03279-6. Online ahead of print.

ABSTRACT

Major depressive disorder (MDD) is a significant neurological disorder that imposes a substantial burden on society, characterized by its high recurrence rate and associated suicide risk. Clinical diagnosis, which relies on interviews with psychiatrists and questionnaires used as auxiliary diagnostic tools, lacks precision and objectivity in diagnosing MDD. To address these challenges, this study proposes an assessment method based on EEG. It involves calculating the phase lag index (PLI) in alpha and gamma bands to construct functional brain connectivity. This method aims to find biomarkers to assess the severity of MDD and suicidal ideation. The convolutional inception with shuffled attention network (CISANET) was introduced for this purpose. The study included 61 patients with MDD, who were classified into mild, moderate, and severe levels based on depression scales, and the presence of suicidal ideation was evaluated. Two paradigms were designed for the study, with EEG analysis focusing on 32 selected electrodes to extract alpha and gamma bands. In the gamma band, the classification accuracy reached 77.37% in the visual paradigm and 80.12% in the auditory paradigm. The average accuracy in classifying suicidal ideation was 93.60%. The findings suggest that gamma bands can be used as potential biomarkers differentiating illness severity and identifying suicidal ideation of MDD, and that objective assessment methods can effectively assess MDD The objective assessment method can effectively assess the severity of MDD and identify suicidal ideation of MDD patients, which provides a valuable theoretical basis for understanding the biological characteristics of MDD.

PMID:39849234 | DOI:10.1007/s11517-024-03279-6

J Imaging Inform Med. 2025 Jan 23. doi: 10.1007/s10278-025-01402-z. Online ahead of print.

ABSTRACT

Parkinson's disease (PD), a degenerative disorder of the central nervous system, is commonly diagnosed using functional medical imaging techniques such as single-photon emission computed tomography (SPECT). In this study, we utilized two SPECT data sets (n = 634 and n = 202) from different hospitals to develop a model capable of accurately predicting PD stages, a multiclass classification task. We used the entire three-dimensional (3D) brain images as input and experimented with various model architectures. Initially, we treated the 3D images as sequences of two-dimensional (2D) slices and fed them sequentially into 2D convolutional neural network (CNN) models pretrained on ImageNet, averaging the outputs to obtain the final predicted stage. We also applied 3D CNN models pretrained on Kinetics-400. Additionally, we incorporated an attention mechanism to account for the varying importance of different slices in the prediction process. To further enhance model efficacy and robustness, we simultaneously trained the two data sets using weight sharing, a technique known as cotraining. Our results demonstrated that 2D models pretrained on ImageNet outperformed 3D models pretrained on Kinetics-400, and models utilizing the attention mechanism outperformed both 2D and 3D models. The cotraining technique proved effective in improving model performance when the cotraining data sets were sufficiently large.

PMID:39849204 | DOI:10.1007/s10278-025-01402-z

J Imaging Inform Med. 2025 Jan 23. doi: 10.1007/s10278-025-01386-w. Online ahead of print.

ABSTRACT

Accurate wound segmentation is crucial for the precise diagnosis and treatment of various skin conditions through image analysis. In this paper, we introduce a novel dual attention U-Net model designed for precise wound segmentation. Our proposed architecture integrates two widely used deep learning models, VGG16 and U-Net, incorporating dual attention mechanisms to focus on relevant regions within the wound area. Initially trained on diabetic foot ulcer images, we fine-tuned the model to acute and chronic wound images and conducted a comprehensive comparison with other state-of-the-art models. The results highlight the superior performance of our proposed dual attention model, achieving a Dice coefficient and IoU of 94.1% and 89.3%, respectively, on the test set. This underscores the robustness of our method and its capacity to generalize effectively to new data.

PMID:39849203 | DOI:10.1007/s10278-025-01386-w

J Imaging Inform Med. 2025 Jan 23. doi: 10.1007/s10278-025-01390-0. Online ahead of print.

ABSTRACT

While radiation hazards induced by cone-beam computed tomography (CBCT) in image-guided radiotherapy (IGRT) can be reduced by sparse-view sampling, the image quality is inevitably degraded. We propose a deep learning-based multi-view projection synthesis (DLMPS) approach to improve the quality of sparse-view low-dose CBCT images. In the proposed DLMPS approach, linear interpolation was first applied to sparse-view projections and the projections were rearranged into sinograms; these sinograms were processed with a sinogram restoration model and then rearranged back into projections. The sinogram restoration model was modified from the 2D U-Net by incorporating dynamic convolutional layers and residual learning techniques. The DLMPS approach was trained, validated, and tested on CBCT data from 163, 30, and 30 real patients respectively. Sparse-view projection datasets with 1/4 and 1/8 of the original sampling rate were simulated, and the corresponding full-view projection datasets were restored via the DLMPS approach. Tomographic images were reconstructed using the Feldkamp-Davis-Kress algorithm. Quantitative metrics including root-mean-square error (RMSE), peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and feature similarity (FSIM) were calculated in both the projection and image domains to evaluate the performance of the DLMPS approach. The DLMPS approach was compared with 11 state-of-the-art (SOTA) models, including CNN and Transformer architectures. For 1/4 sparse-view reconstruction task, the proposed DLMPS approach achieved averaged RMSE, PSNR, SSIM, and FSIM values of 0.0271, 45.93 dB, 0.9817, and 0.9587 in the projection domain, and 0.000885, 37.63 dB, 0.9074, and 0.9885 in the image domain, respectively. For 1/8 sparse-view reconstruction task, the DLMPS approach achieved averaged RMSE, PSNR, SSIM, and FSIM values of 0.0304, 44.85 dB, 0.9785, and 0.9524 in the projection domain, and 0.001057, 36.05 dB, 0.8786, and 0.9774 in the image domain, respectively. The DLMPS approach outperformed all the 11 SOTA models in both the projection and image domains for 1/4 and 1/8 sparse-view reconstruction tasks. The proposed DLMPS approach effectively improves the quality of sparse-view CBCT images in IGRT by accurately synthesizing missing projections, exhibiting potential in substantially reducing imaging dose to patients with minimal loss of image quality.

PMID:39849201 | DOI:10.1007/s10278-025-01390-0