Deep learning
Inhibition of tumour necrosis factor alpha by Etanercept attenuates Shiga toxin-induced brain pathology
J Neuroinflammation. 2025 Feb 7;22(1):33. doi: 10.1186/s12974-025-03356-z.
ABSTRACT
Infection with enterohemorrhagic E. coli (EHEC) causes severe changes in the brain leading to angiopathy, encephalopathy and microglial activation. In this study, we investigated the role of tumour necrosis factor alpha (TNF-α) for microglial activation and brain pathology using a preclinical mouse model of EHEC infection. LC-MS/MS proteomics of mice injected with a combination of Shiga toxin (Stx) and lipopolysaccharide (LPS) revealed extensive alterations of the brain proteome, in particular enrichment of pathways involved in complement activation and coagulation cascades. Inhibition of TNF-α by the drug Etanercept strongly mitigated these changes, particularly within the complement pathway, suggesting TNF-α-dependent vasodilation and endothelial injury. Analysis of microglial populations using a novel human-in-the-loop deep learning algorithm for the segmentation of microscopic imaging data indicated specific morphological changes, which were reduced to healthy condition after inhibition of TNF-α. Moreover, the Stx/LPS-mediated angiopathy was significantly attenuated by inhibition of TNF-α. Overall, our findings elucidate the critical role of TNF-α in EHEC-induced brain pathology and highlight a potential therapeutic target for mitigating neuroinflammation, microglial activation and injury associated with EHEC infection.
PMID:39920757 | DOI:10.1186/s12974-025-03356-z
Predicting hematoma expansion after intracerebral hemorrhage: a comparison of clinician prediction with deep learning radiomics models
Neurocrit Care. 2025 Feb 7. doi: 10.1007/s12028-025-02214-3. Online ahead of print.
ABSTRACT
BACKGROUND: Early prediction of hematoma expansion (HE) following nontraumatic intracerebral hemorrhage (ICH) may inform preemptive therapeutic interventions. We sought to identify how accurately machine learning (ML) radiomics models predict HE compared with expert clinicians using head computed tomography (HCT).
METHODS: We used data from 900 study participants with ICH enrolled in the Antihypertensive Treatment of Acute Cerebral Hemorrhage 2 Study. ML models were developed using baseline HCT images, as well as admission clinical data in a training cohort (n = 621), and their performance was evaluated in an independent test cohort (n = 279) to predict HE (defined as HE by 33% or > 6 mL at 24 h). We simultaneously surveyed expert clinicians and asked them to predict HE using the same initial HCT images and clinical data. Area under the receiver operating characteristic curve (AUC) were compared between clinician predictions, ML models using radiomic data only (a random forest classifier and a deep learning imaging model) and ML models using both radiomic and clinical data (three random forest classifier models using different feature combinations). Kappa values comparing interrater reliability among expert clinicians were calculated. The best performing model was compared with clinical predication.
RESULTS: The AUC for expert clinician prediction of HE was 0.591, with a kappa of 0.156 for interrater variability, compared with ML models using radiomic data only (a deep learning model using image input, AUC 0.680) and using both radiomic and clinical data (a random forest model, AUC 0.677). The intraclass correlation coefficient for clinical judgment and the best performing ML model was 0.47 (95% confidence interval 0.23-0.75).
CONCLUSIONS: We introduced supervised ML algorithms demonstrating that HE prediction may outperform practicing clinicians. Despite overall moderate AUCs, our results set a new relative benchmark for performance in these tasks that even expert clinicians find challenging. These results emphasize the need for continued improvements and further enhanced clinical decision support to optimally manage patients with ICH.
PMID:39920546 | DOI:10.1007/s12028-025-02214-3
A generative whole-brain segmentation model for positron emission tomography images
EJNMMI Phys. 2025 Feb 8;12(1):15. doi: 10.1186/s40658-025-00716-9.
ABSTRACT
PURPOSE: Whole-brain segmentation via positron emission tomography (PET) imaging is crucial for advancing neuroscience research and clinical medicine, providing essential insights into biological metabolism and activity within different brain regions. However, the low resolution of PET images may have limited the segmentation accuracy of multiple brain structures. Therefore, we propose a generative multi-object segmentation model for brain PET images to achieve automatic and accurate segmentation.
METHODS: In this study, we propose a generative multi-object segmentation model for brain PET images with two learning protocols. First, we pretrained a latent mapping model to learn the mapping relationship between PET and MR images so that we could extract anatomical information of the brain. A 3D multi-object segmentation model was subsequently proposed to apply whole-brain segmentation to MR images generated from integrated latent mapping models. Moreover, a custom cross-attention module based on a cross-attention mechanism was constructed to effectively fuse the functional information and structural information. The proposed method was compared with various deep learning-based approaches in terms of the Dice similarity coefficient, Jaccard index, precision, and recall serving as evaluation metrics.
RESULTS: Experiments were conducted on real brain PET/MR images from 120 patients. Both visual and quantitative results indicate that our method outperforms the other comparison approaches, achieving 75.53% ± 4.26% Dice, 66.02% ± 4.55% Jaccard, 74.64% ± 4.15% recall and 81.40% ± 2.30% precision. Furthermore, the evaluation of the SUV distribution and correlation assessment in the regions of interest demonstrated consistency with the ground truth. Additionally, clinical tolerance rates, which are determined by the tumor background ratio, have confirmed the ability of the method to distinguish highly metabolic regions accurately from normal regions, reinforcing its clinical applicability.
CONCLUSION: For automatic and accurate whole-brain segmentation, we propose a novel 3D generative multi-object segmentation model for brain PET images, which achieves superior model performance compared with other deep learning methods. In the future, we will apply our whole-brain segmentation method to clinical practice and extend it to other multimodal tasks.
PMID:39920478 | DOI:10.1186/s40658-025-00716-9
Voice analysis and deep learning for detecting mental disorders in pregnant women: a cross-sectional study
Discov Ment Health. 2025 Feb 8;5(1):12. doi: 10.1007/s44192-025-00138-0.
ABSTRACT
INTRODUCTION: Perinatal mental disorders are prevalent, affecting 10-20% of pregnant women, and can negatively impact both maternal and neonatal outcomes. Traditional screening tools, such as the Edinburgh Postnatal Depression Scale (EPDS), present limitations due to subjectivity and time constraints in clinical settings. Recent advances in voice analysis and machine learning have shown potential for providing more objective screening methods. This study aimed to develop a deep learning model that analyzes the voices of pregnant women to screen for mental disorders, thereby offering an alternative to the traditional tools.
METHODS: A cross-sectional study was conducted among 204 pregnant women, from whom voice samples were collected during their one-month postpartum checkup. The audio data were preprocessed into 5000 ms intervals, converted into mel-spectrograms, and augmented using TrivialAugment and context-rich minority oversampling. The EfficientFormer V2-L model, pretrained on ImageNet, was employed with transfer learning for classification. The hyperparameters were optimized using Optuna, and an ensemble learning approach was used for the final predictions. The model's performance was compared to that of the EPDS in terms of sensitivity, specificity, and other diagnostic metrics.
RESULTS: Of the 172 participants analyzed (149 without mental disorders and 23 with mental disorders), the voice-based model demonstrated a sensitivity of 1.00 and a recall of 0.82, outperforming the EPDS in these areas. However, the EPDS exhibited higher specificity (0.97) and precision (0.84). No significant difference was observed in the area under the receiver operating characteristic curve between the two methods (p = 0.759).
DISCUSSION: The voice-based model showed higher sensitivity and recall, suggesting that it may be more effective in identifying at-risk individuals than the EPDS. Machine learning and voice analysis are promising objective screening methods for mental disorders during pregnancy, potentially improving early detection.
CONCLUSION: We developed a lightweight machine learning model to analyze pregnant women's voices for screening various mental disorders, achieving high sensitivity and demonstrating the potential of voice analysis as an effective and objective tool in perinatal mental health care.
PMID:39920468 | DOI:10.1007/s44192-025-00138-0
A deep learning-driven method for safe and effective ERCP cannulation
Int J Comput Assist Radiol Surg. 2025 Feb 7. doi: 10.1007/s11548-025-03329-w. Online ahead of print.
ABSTRACT
PURPOSE: In recent years, the detection of the duodenal papilla and surgical cannula has become a critical task in computer-assisted endoscopic retrograde cholangiopancreatography (ERCP) cannulation operations. The complex surgical anatomy, coupled with the small size of the duodenal papillary orifice and its high similarity to the background, poses significant challenges to effective computer-assisted cannulation. To address these challenges, we present a deep learning-driven graphical user interface (GUI) to assist ERCP cannulation.
METHODS: Considering the characteristics of the ERCP scenario, we propose a deep learning method for duodenal papilla and surgical cannula detection, utilizing four swin transformer decoupled heads (4STDH). Four different prediction heads are employed to detect objects of different sizes. Subsequently, we integrate the swin transformer module to identify attention regions to explore prediction potential deeply. Moreover, we decouple the classification and regression networks, significantly improving the model's accuracy and robustness through the separation prediction. Simultaneously, we introduce a dataset on papilla and cannula (DPAC), consisting of 1840 annotated endoscopic images, which will be publicly available. We integrated 4STDH and several state-of-the-art methods into the GUI and compared them.
RESULTS: On the DPAC dataset, 4STDH outperforms state-of-the-art methods with an mAP of 93.2% and superior generalization performance. Additionally, the GUI provides real-time positions of the papilla and cannula, along with the planar distance and direction required for the cannula to reach the cannulation position.
CONCLUSION: We validate the GUI's performance in human gastrointestinal endoscopic videos, showing deep learning's potential to enhance the safety and efficiency of clinical ERCP cannulation.
PMID:39920403 | DOI:10.1007/s11548-025-03329-w
Deep learning-based multimodal integration of imaging and clinical data for predicting surgical approach in percutaneous transforaminal endoscopic discectomy
Eur Spine J. 2025 Feb 8. doi: 10.1007/s00586-025-08668-5. Online ahead of print.
ABSTRACT
BACKGROUND: For cases of multilevel lumbar disc herniation (LDH), selecting the surgical approach for Percutaneous Transforaminal Endoscopic Discectomy (PTED) presents significant challenges and heavily relies on the physician's judgment. This study aims to develop a deep learning (DL)-based multimodal model that provides objective and referenceable support by comprehensively analyzing imaging and clinical data to assist physicians.
METHODS: This retrospective study collected imaging and clinical data from patients with multilevel LDH. Each segmental MR scan was concurrently fed into a multi-input ResNet 50 model to predict the target segment. The target segment scan was then input to a custom model to predict the PTED approach direction. Clinical data, including the patient's lower limb sensory and motor functions, were used as feature variables in a machine learning (ML) model for prediction. Bayesian optimization was employed to determine the optimal weights for the fusion of the two models.
RESULT: The predictive performance of the multimodal model significantly outperformed the DL and ML models. For PTED target segment prediction, the multimodal model achieved an accuracy of 93.8%, while the DL and ML models achieved accuracies of 87.7% and 87.0%, respectively. Regarding the PTED approach direction, the multimodal model had an accuracy of 89.3%, significantly higher than the DL model's 87.8% and the ML model's 87.6%.
CONCLUSION: The multimodal model demonstrated excellent performance in predicting PTED target segments and approach directions. Its predictive performance surpassed that of the individual DL and ML models.
PMID:39920320 | DOI:10.1007/s00586-025-08668-5
Deep learning radiomics model based on contrast-enhanced MRI for distinguishing between tuberculous spondylitis and pyogenic spondylitis
Eur Spine J. 2025 Feb 8. doi: 10.1007/s00586-025-08696-1. Online ahead of print.
ABSTRACT
PURPOSE: This study aimed to develop and validate a deep learning radiomics nomogram (DLRN) to differentiate between tuberculous spondylitis (TS) and pyogenic spondylitis (PS) using contrast-enhanced MRI (CE-MRI).
METHODS: A retrospective approach was employed, enrolling patients diagnosed with TS or PS based on pathological examination at two centers. Clinical features were evaluated to establish a clinical model. Radiomics and deep learning (DL) features were extracted from contrast-enhanced T1-weighted images and subsequently fused. Following feature selection, radiomics, DL, combined DL-radiomics (DLR), and a deep learning radiomics nomogram (DLRN) were developed to differentiate TS from PS. Performance was assessed using metrics including the area under the curve (AUC), calibration curves, and decision curve analysis (DCA).
RESULTS: A total of 147 patients met the study criteria. Center 1 comprised the training cohort with 102 patients (52 TS and 50 PS), while Center 2 served as the external test cohort with 45 patients (17 TS and 28 PS). The DLRN model exhibited the highest diagnostic accuracy, achieving an AUC of 0.994 (95% CI: 0.983-1.000) in the training cohort and 0.859 (95% CI: 0.744-0.975) in the external test cohort. Calibration curves indicated good agreement for DLRN, and decision curve analysis (DCA) demonstrated it provided the greatest clinical benefit.
CONCLUSION: The CE-MRI-based DLRN showed robust diagnostic capability for distinguishing between TS and PS in clinical practice.
PMID:39920318 | DOI:10.1007/s00586-025-08696-1
FoxA1 knockdown promotes BMSC osteogenesis in part by activating the ERK1/2 signaling pathway and preventing ovariectomy-induced bone loss
Sci Rep. 2025 Feb 7;15(1):4594. doi: 10.1038/s41598-025-88658-1.
ABSTRACT
The influence of deep learning in the medical and molecular biology sectors is swiftly growing and holds the potential to improve numerous crucial domains. Osteoporosis is a significant global health issue, and the current treatment options are highly restricted. Transplanting genetically engineered MSCs has been acknowledged as a highly promising therapy for osteoporosis. We utilized a random walk-based technique to discern genes associated with ossification. The osteogenic value of these genes was assessed on the basis of information found in published scientific literature. GO enrichment analysis of these genes was performed to determine if they were enriched in any certain function. Immunohistochemical and western blot techniques were used to identify and measure protein expression. The expression of genes involved in osteogenic differentiation was examined via qRT‒PCR. Lentiviral transfection was utilized to suppress the expression of the FOXA1 gene in hBMSCs. An in vivo mouse model of ovariectomy was created, and radiographic examination was conducted to confirm the impact of FOXA1 knockdown on osteoporosis. The osteogenic score of each gene was calculated by assessing its similarity to osteo-specific genes. The majority of the genes with the highest rankings were linked with osteogenic differentiation, indicating that our approach is useful for identifying genes associated with ossification. GO enrichment analysis revealed that these pathways are enriched primarily in bone-related processes. FOXA1 is a crucial transcription factor that controls the process of osteogenic differentiation, as indicated by similarity analysis. FOXA1 was significantly increased in those with osteoporosis. Downregulation of FOXA1 markedly augmented the expression of osteoblast-specific genes and proteins, activated the ERK1/2 signaling pathway, intensified ALP activity, and promoted mineral deposition. In addition, excessive expression of FOXA1 significantly reduced ALP activity and mineral deposits. Using a mouse model in which the ovaries were surgically removed, researchers reported that suppressing the FOXA1 gene in bone marrow stem cells (BMSCs) prevented the loss of bone density caused by ovariectomy. This finding was confirmed by analyzing the bone structure via micro-CT. Furthermore, our approach can distinguish genes that exhibit osteogenic differentiation characteristics. This ability can aid in the identification of novel genes associated with osteogenic differentiation, which can be utilized in the treatment of osteoporosis. Computational and laboratory evidence indicates that reducing the expression of FOXA1 enhances the process of bone formation in bone marrow-derived mesenchymal stem cells (BMSCs) and may serve as a promising approach to prevent osteoporosis.
PMID:39920313 | DOI:10.1038/s41598-025-88658-1
A deep-learning approach to parameter fitting for a lithium metal battery cycling model, validated with experimental cell cycling time series
Sci Rep. 2025 Feb 7;15(1):4620. doi: 10.1038/s41598-025-87830-x.
ABSTRACT
Symmetric coin cell cycling is an important tool for the analysis of battery materials, enabling the study of electrode/electrolyte systems under realistic operating conditions. In the case of metal lithium SEI growth and shape changes, cycling studies are especially important to assess the impact of the alternation of anodic-cathodic polarization with the relevant electrolyte geometry and mass-transport conditions. Notwithstanding notable progress in analysis of lithium/lithium symmetric coin cell cycling data, on the one hand, some aspects of the cell electrochemical response still warrant investigation, and, on the other hand, very limited quantitative use is made of large corpora of experimental data generated in electrochemical experiments. This study contributes to shedding light on this highly technologically relevant problem, thanks to the combination of quantitative data exploitation and Partial Differential Equation (PDE) modelling for metal anode cycling. Toward this goal, we propose the use of a Convolutional Neural Network-Long-Short Term Memory (CNN-LSTM) to identify relevant physico-chemical parameters in the PDE model and to describe the behaviour of simulated and experimental charge-discharge profiles. Specifically, we have carried out parameter identification tasks for experimental data regarding the cycling of symmetric coin cells with Li chips as electrodes and LP30 electrolyte. Representative selection of numerical results highlights the advantages of this new approach with respect to traditional Least Squares fitting.
PMID:39920238 | DOI:10.1038/s41598-025-87830-x
Advanced retinal disease detection from OCT images using a hybrid squeeze and excitation enhanced model
PLoS One. 2025 Feb 7;20(2):e0318657. doi: 10.1371/journal.pone.0318657. eCollection 2025.
ABSTRACT
BACKGROUND: Retinal problems are critical because they can cause severe vision loss if not treated. Traditional methods for diagnosing retinal disorders often rely heavily on manual interpretation of optical coherence tomography (OCT) images, which can be time-consuming and dependent on the expertise of ophthalmologists. This leads to challenges in early diagnosis, especially as retinal diseases like diabetic macular edema (DME), Drusen, and Choroidal neovascularization (CNV) become more prevalent. OCT helps ophthalmologists diagnose patients more accurately by allowing for early detection. This paper offers a hybrid SE (Squeeze-and-Excitation)-Enhanced Hybrid Model for detecting retinal disorders from OCT images, including DME, Drusen, and CNV, using artificial intelligence and deep learning.
METHODS: The model integrates SE blocks with EfficientNetB0 and Xception architectures, which provide high success in image classification tasks. EfficientNetB0 achieves high accuracy with fewer parameters through model scaling strategies, while Xception offers powerful feature extraction using deep separable convolutions. The combination of these architectures enhances both the efficiency and classification performance of the model, enabling more accurate detection of retinal disorders from OCT images. Additionally, SE blocks increase the representational ability of the network by adaptively recalibrating per-channel feature responses.
RESULTS: The combined features from EfficientNetB0 and Xception are processed via fully connected layers and categorized using the Softmax algorithm. The methodology was tested on UCSD and Duke's OCT datasets and produced excellent results. The proposed SE-Improved Hybrid Model outperformed the current best-known approaches, with accuracy rates of 99.58% on the UCSD dataset and 99.18% on the Duke dataset.
CONCLUSION: These findings emphasize the model's ability to effectively diagnose retinal disorders using OCT images and indicate substantial promise for the development of computer-aided diagnostic tools in the field of ophthalmology.
PMID:39919140 | DOI:10.1371/journal.pone.0318657
Multi-dimensional perceptual recognition of tourist destination using deep learning model and geographic information system
PLoS One. 2025 Feb 7;20(2):e0318846. doi: 10.1371/journal.pone.0318846. eCollection 2025.
ABSTRACT
Perceptual recognition of tourist destinations is vital in representing the destination image, supporting destination management decision-making, and promoting tourism recommendations. However, previous studies on tourist destination perception have limitations regarding accuracy and completeness related to research methods. This study addresses these limitations by proposing an efficient strategy to achieve precise perceptual recognition of tourist destinations while ensuring the integrity of user-generated content (UGC) data and the completeness of perception dimensions. We integrated various types of UGC data, including images, texts, and spatiotemporal information, to create a comprehensive UGC dataset. Then, we adopted the improved Inception V3 model, the bidirectional long short-term memory network (BiLSTM) model with multi-head attention, and geographic information system (GIS) technology to recognize basic tourist feature information from the UGC dataset, such as the content, sentiment, and spatiotemporal perceptual dimensions of the data, achieving a recognition accuracy of over 97%. Finally, a progressive dimension combination method was proposed to visualize and analyze multiple perceptions. An experimental case study demonstrated the strategy's effectiveness, focusing on tourists' perceptions of Datong, China. Experimental results show that the approach is feasible for studying tourist destination perception. Content perception, sentiment perception, and the perception of Datong's spatial and temporal characteristics were recognized and analyzed efficiently. This study offers valuable guidance and a reference framework for selecting methods and technical routes in tourist destination perception.
PMID:39919101 | DOI:10.1371/journal.pone.0318846
scCamAge: A context-aware prediction engine for cellular age, aging-associated bioactivities, and morphometrics
Cell Rep. 2025 Feb 6;44(2):115270. doi: 10.1016/j.celrep.2025.115270. Online ahead of print.
ABSTRACT
Current deep-learning-based image-analysis solutions exhibit limitations in holistically capturing spatiotemporal cellular changes, particularly during aging. We present scCamAge, an advanced context-aware multimodal prediction engine that co-leverages image-based cellular spatiotemporal features at single-cell resolution alongside cellular morphometrics and aging-associated bioactivities such as genomic instability, mitochondrial dysfunction, vacuolar dynamics, reactive oxygen species levels, and epigenetic and proteasomal dysfunctions. scCamAge employed heterogeneous datasets comprising ∼1 million single yeast cells and was validated using pro-longevity drugs, genetic mutants, and stress-induced models. scCamAge also predicted a pro-longevity response in yeast cells under iterative thermal stress, confirmed using integrative omics analyses. Interestingly, scCamAge, trained solely on yeast images, without additional learning, surpasses generic models in predicting chemical and replication-induced senescence in human fibroblasts, indicating evolutionary conservation of aging-related morphometrics. Finally, we enhanced the generalizability of scCamAge by retraining it on human fibroblast senescence datasets, which improved its ability to predict senescent cells.
PMID:39918957 | DOI:10.1016/j.celrep.2025.115270
A novel deep learning framework for retinal disease detection leveraging contextual and local features cues from retinal images
Med Biol Eng Comput. 2025 Feb 7. doi: 10.1007/s11517-025-03314-0. Online ahead of print.
ABSTRACT
Retinal diseases are a serious global threat to human vision, and early identification is essential for effective prevention and treatment. However, current diagnostic methods rely on manual analysis of fundus images, which heavily depends on the expertise of ophthalmologists. This manual process is time-consuming and labor-intensive and can sometimes lead to missed diagnoses. With advancements in computer vision technology, several automated models have been proposed to improve diagnostic accuracy for retinal diseases and medical imaging in general. However, these methods face challenges in accurately detecting specific diseases within images due to inherent issues associated with fundus images, including inter-class similarities, intra-class variations, limited local information, insufficient contextual understanding, and class imbalances within datasets. To address these challenges, we propose a novel deep learning framework for accurate retinal disease classification. This framework is designed to achieve high accuracy in identifying various retinal diseases while overcoming inherent challenges associated with fundus images. Generally, the framework consists of three main modules. The first module is Densely Connected Multidilated Convolution Neural Network (DCM-CNN) that extracts global contextual information by effectively integrating novel Casual Dilated Dense Convolutional Blocks (CDDCBs). The second module of the framework, namely, Local-Patch-based Convolution Neural Network (LP-CNN), utilizes Class Activation Map (CAM) (obtained from DCM-CNN) to extract local and fine-grained information. To identify the correct class and minimize the error, a synergic network is utilized that takes the feature maps of both DCM-CNN and LP-CNN and connects both maps in a fully connected fashion to identify the correct class and minimize the errors. The framework is evaluated through a comprehensive set of experiments, both quantitatively and qualitatively, using two publicly available benchmark datasets: RFMiD and ODIR-5K. Our experimental results demonstrate the effectiveness of the proposed framework and achieves higher performance on RFMiD and ODIR-5K datasets compared to reference methods.
PMID:39918766 | DOI:10.1007/s11517-025-03314-0
Evaluation of deep learning-based scatter correction on a long-axial field-of-view PET scanner
Eur J Nucl Med Mol Imaging. 2025 Feb 7. doi: 10.1007/s00259-025-07120-6. Online ahead of print.
ABSTRACT
OBJECTIVE: Long-axial field-of-view (LAFOV) positron emission tomography (PET) systems allow higher sensitivity, with an increased number of detected lines of response induced by a larger angle of acceptance. However this extended angle increases the number of multiple scatters and the scatter contribution within oblique planes. As scattering affects both quality and quantification of the reconstructed image, it is crucial to correct this effect with more accurate methods than the state-of-the-art single scatter simulation (SSS) that can reach its limits with such an extended field-of-view (FOV). In this work, which is an extension of our previous assessment of deep learning-based scatter estimation (DLSE) carried out on a conventional PET system, we aim to evaluate the DLSE method performance on LAFOV total-body PET.
APPROACH: The proposed DLSE method based on an convolutional neural network (CNN) U-Net architecture uses emission and attenuation sinograms to estimate scatter sinogram. The network was trained from Monte-Carlo (MC) simulations of XCAT phantoms [ 18 F]-FDG PET acquisitions using a Siemens Biograph Vision Quadra scanner model, with multiple morphologies and dose distributions. We firstly evaluated the method performance on simulated data in both sinogram and image domain by comparing it to the MC ground truth and SSS scatter sinograms. We then tested the method on seven [ 18 F]-FDG and [ 18 F]-PSMA clinical datasets, and compare it to SSS estimations.
RESULTS: DLSE showed superior accuracy on phantom data, greater robustness to patient size and dose variations compared to SSS, and better lesion contrast recovery. It also yielded promising clinical results, improving lesion contrasts in [ 18 F]-FDG datasets and performing consistently with [ 18 F]-PSMA datasets despite no training with [ 18 F]-PSMA.
SIGNIFICANCE: LAFOV PET scatter can be accurately estimated from raw data using the proposed DLSE method.
PMID:39918764 | DOI:10.1007/s00259-025-07120-6
Applying deep learning for underwater broadband-source detection using a spherical array
J Acoust Soc Am. 2025 Feb 1;157(2):947-961. doi: 10.1121/10.0035787.
ABSTRACT
For improving passive detection of underwater broadband sources, a source-detection and direction-of-arrival-estimation method is developed herein based on a deep neural network (DNN) using a spherical array. Spherical Fourier transform is employed to convert the element pressure signals into spherical Fourier coefficients, which are used as inputs of the DNN. A Gaussian distribution with a spatial-spectrum-like form is adopted to design labels for the DNN. A physical model coupling underwater acoustic propagation and the spherical array is established to simulate array signals for DNN training. The introduction of white noise into the training data considerably enhances the detection capability of the DNN and effectively suppresses false estimation. The model's performance is evaluated based on its detection rate at a constant false alarm rate. Notably, the model does not rely on prior knowledge of the source's spectral features. Further, this study demonstrates that a DNN trained by one source can achieve multisource detection to a certain extent. The simulation and experimental processing results validate the broadband detection capability of the proposed method at varying signal-to-noise ratios.
PMID:39918577 | DOI:10.1121/10.0035787
Automated Description Generation of Cytologic Findings for Lung Cytological Images Using a Pretrained Vision Model and Dual Text Decoders: Preliminary Study
Cytopathology. 2025 Feb 7. doi: 10.1111/cyt.13474. Online ahead of print.
ABSTRACT
OBJECTIVE: Cytology plays a crucial role in lung cancer diagnosis. Pulmonary cytology involves cell morphological characterisation in the specimen and reporting the corresponding findings, which are extremely burdensome tasks. In this study, we propose a technique to generate cytologic findings from for cytologic images to assist in the reporting of pulmonary cytology.
METHODS: For this study, 801 patch images were retrieved using cytology specimens collected from 206 patients; the findings were assigned to each image as a dataset for generating cytologic findings. The proposed method consists of a vision model and dual text decoders. In the former, a convolutional neural network (CNN) is used to classify a given image as benign or malignant, and the features related to the image are extracted from the intermediate layer. Independent text decoders for benign and malignant cells are prepared for text generation, and the text decoder switches according to the CNN classification results. The text decoder is configured using a transformer that uses the features obtained from the CNN for generating findings.
RESULTS: The sensitivity and specificity were 100% and 96.4%, respectively, for automated benign and malignant case classification, and the saliency map indicated characteristic benign and malignant areas. The grammar and style of the generated texts were confirmed correct, achieving a BLEU-4 score of 0.828, reflecting high degree of agreement with the gold standard, outperforming existing LLM-based image-captioning methods and single-text-decoder ablation model.
CONCLUSION: Experimental results indicate that the proposed method is useful for pulmonary cytology classification and generation of cytologic findings.
PMID:39918342 | DOI:10.1111/cyt.13474
I-Brainer: Artificial intelligence/Internet of Things (AI/IoT)-Powered Detection of Brain Cancer
Curr Med Imaging. 2025 Feb 4. doi: 10.2174/0115734056333393250117164020. Online ahead of print.
ABSTRACT
BACKGROUND/OBJECTIVE: Brain tumour is characterized by its aggressive nature and low survival rate and thus regarded as one of the deadliest diseases. Thus, miss-diagnosis or miss-classification of brain tumour can lead to miss treatment or incorrect treatment and reduce survival chances. Therefore, there is need to develop a technique that can identify and detect brain tumour at early stages.
METHODS: Here, we proposed a framework titled I-Brainer which is an Artificial Intelligence/Internet of Things (AI/IoT)-powered classification of MRI. We employed a Br35H+SARTAJ brain MRI dataset which contain 7023 total images which include No tumour, pituitary, meningioma and glioma. In order to accurately classified MRI into 4-class, we developed LeNet model from scratch, implemented 2 pretrained models which include EfficientNet and ResNet-50 as well feature extraction of these models coupled with 2 Machine Learning classifiers k-Nearest Neighbours (KNN) and Support Vector Machines (SVM).
RESULT: Evaluation and comparison of the performance of 3 models has shown that EfficientNet+SVM achieved the best result in terms of AUC (99%) and ResNet-50-KNN ranked higher in terms of accuracy (94%) on testing dataset.
CONCLUSION: This framework can be harness by patients residing in remote areas and as confirmatory approach for medical experts.
PMID:39917913 | DOI:10.2174/0115734056333393250117164020
Design and structure of overlapping regions in PCA via deep learning
Synth Syst Biotechnol. 2024 Dec 27;10(2):442-451. doi: 10.1016/j.synbio.2024.12.007. eCollection 2025 Jun.
ABSTRACT
Polymerase cycling assembly (PCA) stands out as the predominant method in the synthesis of kilobase-length DNA fragments. The design of overlapping regions is the core factor affecting the success rate of synthesis. However, there still exists DNA sequences that are challenging to design and construct in the genome synthesis. Here we proposed a deep learning model based on extensive synthesis data to discern latent sequence representations in overlapping regions with an AUPR of 0.805. Utilizing the model, we developed the SmartCut algorithm aimed at designing oligonucleotides and enhancing the success rate of PCA experiments. This algorithm was successfully applied to sequences with diverse synthesis constraints, 80.4 % of which were synthesized in a single round. We further discovered structure differences represented by major groove width, stagger, slide, and centroid distance between overlapping and non-overlapping regions, which elucidated the model's reasonableness through the lens of physical chemistry. This comprehensive approach facilitates streamlined and efficient investigations into the genome synthesis.
PMID:39917768 | PMC:PMC11799973 | DOI:10.1016/j.synbio.2024.12.007
Advanced AI-assisted panoramic radiograph analysis for periodontal prognostication and alveolar bone loss detection
Front Dent Med. 2025 Jan 6;5:1509361. doi: 10.3389/fdmed.2024.1509361. eCollection 2024.
ABSTRACT
BACKGROUND: Periodontitis is a chronic inflammatory disease affecting the gingival tissues and supporting structures of the teeth, often leading to tooth loss. The condition begins with the accumulation of dental plaque, which initiates an immune response. Current radiographic methods for assessing alveolar bone loss are subjective, time-consuming, and labor-intensive. This study aims to develop an AI-driven model using Convolutional Neural Networks (CNNs) to accurately assess alveolar bone loss and provide individualized periodontal prognoses from panoramic radiographs.
METHODS: A total of 2,000 panoramic radiographs were collected using the same device, based on the periodontal diagnosis codes from the HOSxP Program. Image enhancement techniques were applied, and an AI model based on YOLOv8 was developed to segment teeth, identify the cemento-enamel junction (CEJ), and assess alveolar bone levels. The model quantified bone loss and classified prognoses for each tooth.
RESULTS: The teeth segmentation model achieved 97% accuracy, 90% sensitivity, 96% specificity, and an F1 score of 0.80. The CEJ and bone level segmentation model showed superior results with 98% accuracy, 100% sensitivity, 98% specificity, and an F1 score of 0.90. These findings confirm the models' effectiveness in analyzing panoramic radiographs for periodontal bone loss detection and prognostication.
CONCLUSION: This AI model offers a state-of-the-art approach for assessing alveolar bone loss and predicting individualized periodontal prognoses. It provides a faster, more accurate, and less labor-intensive alternative to current methods, demonstrating its potential for improving periodontal diagnosis and patient outcomes.
PMID:39917716 | PMC:PMC11797906 | DOI:10.3389/fdmed.2024.1509361
Artificial intelligence in dentistry and dental biomaterials
Front Dent Med. 2024 Dec 23;5:1525505. doi: 10.3389/fdmed.2024.1525505. eCollection 2024.
ABSTRACT
Artificial intelligence (AI) technology is being used in various fields and its use is increasingly expanding in dentistry. The key aspects of AI include machine learning (ML), deep learning (DL), and neural networks (NNs). The aim of this review is to present an overview of AI, its various aspects, and its application in biomedicine, dentistry, and dental biomaterials focusing on restorative dentistry and prosthodontics. AI-based systems can be a complementary tool in diagnosis and treatment planning, result prediction, and patient-centered care. AI software can be used to detect restorations, prosthetic crowns, periodontal bone loss, and root canal segmentation from the periapical radiographs. The integration of AI, digital imaging, and 3D printing can provide more precise, durable, and patient-oriented outcomes. AI can be also used for the automatic segmentation of panoramic radiographs showing normal anatomy of the oral and maxillofacial area. Recent advancement in AI in medical and dental sciences includes multimodal deep learning fusion, speech data detection, and neuromorphic computing. Hence, AI has helped dentists in diagnosis, planning, and aid in providing high-quality dental treatments in less time.
PMID:39917699 | PMC:PMC11797767 | DOI:10.3389/fdmed.2024.1525505