Bone. 2025 May 2:117510. doi: 10.1016/j.bone.2025.117510. Online ahead of print.

ABSTRACT

OBJECTIVES: To establish a precise and efficient diagnostic framework for distinguishing medication-related osteonecrosis of the jaw, radiation-induced osteonecrosis of the jaw, and normal bone tissue, thus enhancing clinical decision-making and enabling targeted therapeutic interventions.

METHODS: Raman spectroscopy was applied to investigate bone mineral composition, organic matrix content, and crystallinity in ninety bone tissue samples (30 MRONJ, 30 ORN, 30 control). Each mandible underwent 10 randomized spectral acquisitions, yielding 900 spectra across 200-2200 cm-1. The raw spectral data were preprocessed using Labspec6 software (Horiba Scientific). Principal component analysis (PCA) and linear discriminant analysis (LDA) were employed for feature extraction and classification. Additionally, a ResNet18 deep learning architecture was employed to enhance diagnostic accuracy. The model's performance was evaluated using precision, recall, and the area under the receiver operating characteristic curve to ensure robustness.

RESULTS: The PCA-LDA integration achieved 90.3 % accuracy in differentiating MRONJ, ORN, and healthy bone， with leave-one-out cross-validation confirming 89.1 % classification robustness. Furthermore, the ResNet18 deep learning model outperformed traditional classification methods, achieving 0.926 ± 0.024 accuracy, 0.924 ± 0.026 precision, 0.926 ± 0.024 recall, and 0.985 ± 0.007 AUROC on the validation set.

SIGNIFICANCE: These findings underscore the significant potential of combining Raman spectroscopy with advanced deep learning techniques as a rapid, noninvasive, and highly reliable diagnostic tool. This approach not only enhances the ability to differentiate between MRONJ and ORN but also offers substantial implications for improving patient management and therapeutic outcomes in clinical practice.

PMID:40320103 | DOI:10.1016/j.bone.2025.117510

Int J Biol Macromol. 2025 May 2:143844. doi: 10.1016/j.ijbiomac.2025.143844. Online ahead of print.

ABSTRACT

The insulin receptor (IR) is a transmembrane protein that controls glucose homeostasis and is highly associated with chronic diseases including cancer and neurological. Traditional experimental methods have provided essential insights into IR structure and function, but they are constrained by time, cost, and scalability. To address these limitations, we present a computational technique for IR prediction based on deep learning and multi-information fusion. First, we built sequence-based training and testing datasets. Second, the compositional, word embedding, and evolutionary features were retrieved using the Weighted-Group Dipeptide Composition (W-GDPC), FastText, and Bi-Block-Position Specific Scoring Matrix (BB-PSSM), respectively. Third, we use compositional, word embedding, and evolutionary features to generate multi-perspective fused features (MPFF). Fourth, the Multiscale Bidirectional Temporal Convolutional Network (MBiTCN) is used to train the model to process features at multiscale and analyze sequences in both forward and backward directions. The proposed approach (IR-MBiTCN) outperforms competing deep learning (DL) and machine learning (ML)-based models on training and testing datasets, achieving 83.50 % and 79.43 % accuracy, respectively. This study represents a pioneering use of computational methodology in IR prediction, providing a scalable, efficient alternative to experimental procedures and paving the way for advances in chronic disease therapy and drug discovery.

PMID:40319974 | DOI:10.1016/j.ijbiomac.2025.143844

Comput Methods Programs Biomed. 2025 Apr 28;267:108814. doi: 10.1016/j.cmpb.2025.108814. Online ahead of print.

ABSTRACT

OBJECTIVE: Artificial intelligence (AI) models are effective for analyzing high-quality slit-lamp images but often face challenges in real-world clinical settings due to image variability. This study aims to develop and evaluate a hybrid AI-based image quality control system to classify slit-lamp images, improving diagnostic accuracy and efficiency, particularly in telemedicine applications.

DESIGN: Cross-sectional study.

METHODS: Our Zhejiang Eye Hospital dataset comprised 2982 slit-lamp images as the internal dataset. Two external datasets were included: 13,554 images from the Aier Guangming Eye Hospital (AGEH) and 9853 images from the First People's Hospital of Aksu District in Xinjiang (FPH of Aksu). We developed a Hybrid Prior-Net (HP-Net), a novel network that combines a ResNet-based classification branch with a prior knowledge branch leveraging Hough circle transform and frequency domain blur detection. The two branches' features are channel-wise concatenated at the fully connected layer, enhancing representational power and improving the network's ability to classify eligible, misaligned, blurred, and underexposed corneal images. Model performance was evaluated using metrics such as accuracy, precision, recall, specificity, and F1-score, and compared against the performance of other deep learning models.

RESULTS: The HP-Net outperformed all other models, achieving an accuracy of 99.03 %, precision of 98.21 %, recall of 95.18 %, specificity of 99.36 %, and an F1-score of 96.54 % in image classification. The results demonstrated that HP-Net was also highly effective in filtering slit-lamp images from the other two datasets, AGEH and FPH of Aksu with accuracies of 97.23 % and 96.97 %, respectively. These results underscore the superior feature extraction and classification capabilities of HP-Net across all evaluated metrics.

CONCLUSIONS: Our AI-based image quality control system offers a robust and efficient solution for classifying corneal images, with significant implications for telemedicine applications. By incorporating slightly blurred but diagnostically usable images into training datasets, the system enhances the reliability and adaptability of AI tools for medical imaging quality control, paving the way for more accurate and efficient diagnostic workflows.

PMID:40319841 | DOI:10.1016/j.cmpb.2025.108814

J Anim Sci. 2025 May 4:skaf136. doi: 10.1093/jas/skaf136. Online ahead of print.

ABSTRACT

This study addresses the challenge of limited data availability in animal science, particularly in modeling complex biological processes such as methane emissions from ruminants. We propose a novel rank-based method for generating synthetic databases with correlated non-normal multivariate distributions aimed at enhancing the accuracy and reliability of predictive modeling tools. Our rank-based approach involves a four-step process: (1) fitting distributions to variables using normal or best-fit non-normal distributions, (2) generating synthetic databases, (3) preserving relationships among variables using Spearman correlations, and (4) cleaning datasets to ensure biological plausibility. We compare this method with copula-based approaches to maintain a pre-established correlation structure. The rank-based method demonstrated superior performance in preserving original distribution moments (mean, variance, skewness, kurtosis) and correlation structures compared to copula-based methods. We generated two synthetic databases (normal and non-normal distributions) and applied random forest (RF) and multiple linear model (LM) regression analyses. RF regression outperformed LM in predicting methane emissions, showing higher R² values (0.927 vs. 0.622) and lower standard errors. However, cross-testing revealed that RF regressions exhibit high specificity to distribution types, underperforming when applied to data with differing distributions. In contrast, LM regressions showed robustness across different distribution types. Our findings highlight the importance of understanding distributional assumptions in regression techniques when generating synthetic databases. The study also underscores the potential of synthetic data in augmenting limited samples, addressing class imbalances, and simulating rare scenarios. While our method effectively preserves descriptive statistical properties, we acknowledge the possibility of introducing artificial (unknown) relationships within subsets of the synthetic database. This research uncovered a practical solution for creating realistic, statistically sound datasets when original data is scarce or sensitive. Its application in predicting methane emissions demonstrates the potential to enhance modeling accuracy in animal science. Future research directions include integrating this approach with deep learning, exploring real-world applications, and developing adaptive machine-learning models for diverse data distributions.

PMID:40319357 | DOI:10.1093/jas/skaf136

Comput Biol Med. 2025 May 2;192(Pt A):110172. doi: 10.1016/j.compbiomed.2025.110172. Online ahead of print.

ABSTRACT

Chronic wounds and associated complications present ever growing burdens for clinics and hospitals world wide. Venous, arterial, diabetic, and pressure wounds are becoming increasingly common globally. These conditions can result in highly debilitating repercussions for those affected, with limb amputations and increased mortality risk resulting from infection becoming more common. New methods to assist clinicians in chronic wound care are therefore vital to maintain high quality care standards. This paper presents an improved HarDNet segmentation architecture which integrates a contrast-eliminating component in the initial layers of the network to enhance feature learning. We also utilise a multi-colour space tensor merging process and adjust the harmonic shape of the convolution blocks to facilitate these additional features. We train our proposed model using wound images from light skinned patients and test the model on two test sets (one set with ground truth, and one without) comprising only darker skinned cases. Subjective ratings are obtained from clinical wound experts with intraclass correlation coefficient used to determine inter-rater reliability. For the dark skin tone test set with ground truth, when comparing the baseline results (DSC=0.6389, IoU=0.5350) with the results for the proposed model (DSC=0.7610, IoU=0.6620) we demonstrate improvements in terms of Dice similarity coefficient (+0.1221) and intersection over union (+0.1270). Measures from the qualitative analysis also indicate improvements in terms of high expert ratings, with improvements of >3% demonstrated when comparing the baseline model with the proposed model. This paper presents the first study to focus on darker skin tones for chronic wound segmentation using models trained only on wound images exhibiting lighter skin. Diabetes is highly prevalent in countries where patients have darker skin tones, highlighting the need for a greater focus on such cases. Additionally, we conduct the largest qualitative study to date for chronic wound segmentation. All source code for this study is available at: https://github.com/mmu-dermatology-research/hardnet-cws.

PMID:40318494 | DOI:10.1016/j.compbiomed.2025.110172

Int Orthod. 2025 May 2;23(4):101010. doi: 10.1016/j.ortho.2025.101010. Online ahead of print.

ABSTRACT

BACKGROUND: Dental age (DA) estimation, crucial for appropriate orthodontic and paediatric treatment planning, traditionally relies on the analysis of developmental stages of teeth. Artificial intelligence (AI) has been increasingly employed for DA estimation through dental radiographs. The current study aimed to systematically review the literature on the application of AI models for radiographic DA estimation among subjects undergoing dental maturation.

MATERIAL AND METHODS: The electronic search was conducted through five databases, namely PubMed, Embase, Scopus, Web of Science, and Google Scholar, in July 2024. The search sought studies relying on AI models for DA estimation based on dental radiographs. Data were analysed using STATA software V.14 and heterogeneity was evaluated using I-squared statistics. A random-effects model was employed for meta-analysis. Publication bias was assessed using a funnel plot, Egger's test, Begg's test, and the trim-and-fill method. Heterogeneity was evaluated with a Galbraith plot, and sensitivity analysis tested robustness.

RESULTS: Thirteen studies were deemed eligible for qualitative synthesis, seven of which were included in the meta-analysis. The mean absolute error varied from 0.6915 to 12.04, with accuracy between 0.404 and 0.959. Sensitivity ranged from 0.42 to 1.00, specificity ranged from 0.8014 to 0.982, and positive predictive value ranged from 0.43 to 0.90. The pooled accuracy of seven studies equalled 0.85 (95% CI: 0.79-0.91).

CONCLUSION: The present findings support the effectiveness of AI models in DA estimation of individuals under 25 years old based on their dental radiographs. However, further studies with larger sample sizes for both test and training datasets are suggested to validate the reliability and clinical applicability of AI in DA estimation.

PMID:40318319 | DOI:10.1016/j.ortho.2025.101010

Sci Rep. 2025 May 3;15(1):15521. doi: 10.1038/s41598-025-99951-4.

ABSTRACT

The rapid increase in Unmanned Aerial Vehicle (UAV) deployments has led to growing concerns about their detection and differentiation from birds, particularly in sensitive areas like airports. Existing detection systems often struggle to distinguish between UAVs and birds due to their similar flight patterns, resulting in high false positive rates and missed detections. This research presents a bio-inspired deep learning model, the Spatiotemporal Bio-Response Neural Network (STBRNN), designed to enhance the differentiation between UAVs and birds in real-time. The model consists of three core components: a Bio-Inspired Convolutional Neural Network (Bio-CNN) for spatial feature extraction, Gated Recurrent Units (GRUs) for capturing temporal motion dynamics, and a novel Bio-Response Layer that adjusts attention based on movement intensity, object proximity, and velocity consistency. The dataset used includes labeled images and videos of UAVs and birds captured in various environments, processed following YOLOv7 specifications. Extensive experiments were conducted comparing STBRNN with five state-of-the-art models, including YOLOv5, Faster R-CNN, SSD, RetinaNet, and R-FCN. The results demonstrate that STBRNN achieves superior performance across multiple metrics, with a precision of 0.984, recall of 0.964, F1 score of 0.974, and an IoU of 0.96. Additionally, STBRNN operates at an inference time of 45ms per frame, making it highly suitable for real-time applications in UAV and bird detection.

PMID:40319117 | DOI:10.1038/s41598-025-99951-4

Sci Rep. 2025 May 3;15(1):15539. doi: 10.1038/s41598-025-95729-w.

ABSTRACT

Aiming at the student-centered online music learning platform, this study proposes a Course Recommendation Model for Student Learning Interest Evolution (CRM-SLIE) to improve the accuracy and adaptability of the platform's course recommendation. This model combines attention mechanism and Gated Recurrent Unit (GRU), and introduces project crossing module, which can effectively capture students' interest changes and second-order characteristic interaction among courses. The experimental results show that the CRM-SLIE model has excellent performance under different embedding dimensions and the length of student behavior sequence. Especially when the embedding dimension is 64, the Area Under the Curve (AUC) of the model is the highest, and the performance tends to be stable when the sequence length is 20, which is 0.872. Further recall experiments show that with the increase of the number of recommendations, the highest recall rate of CRM-SLIE is 0.364, which is better than other comparative models and can better meet the learning needs of students. In addition, the results of ablation experiments show that the position coding and the way of item crossing have a significant impact on the model performance, and the combination of inner product and Hadamard product is particularly effective in capturing the complex relationship among courses. The research shows that CRM-SLIE model has strong adaptability, robustness and practical application value in the course recommendation task, and can provide personalized and accurate learning resource recommendation for online music learning platform.

PMID:40319107 | DOI:10.1038/s41598-025-95729-w

Sci Rep. 2025 May 3;15(1):15543. doi: 10.1038/s41598-025-00309-7.

ABSTRACT

With advancements in technology, lung nodule detection has significantly improved in both speed and accuracy. However, challenges remain in deploying these methods in complex real-world scenarios. This paper introduces an enhanced lung nodule detection algorithm base on RT-DETR, called LN-DETR. First, we designed a Deep and Shallow Detail Fusion layer that effectively fuses cross-scale features from both shallow and deep layers. Second, we optimized the computational load of the backbone network, effectively reducing the overall scale of the model. Finally, an efficient downsampling is designed to enhance the detection of lung nodules by re-weighting contextual information. Experiments conducted on the public LUNA16 dataset demonstrate that the proposed method, with a reduced number of parameters and computational overhead, achieves 83.7% mAP@0.5 and 36.3% mAP@0.5:0.95, outperforming RT-DETR in both model size and accuracy. These results highlight the superior detection accuracy of the proposed network while maintaining computational efficiency.

PMID:40319047 | DOI:10.1038/s41598-025-00309-7

J Voice. 2025 May 2:S0892-1997(25)00137-7. doi: 10.1016/j.jvoice.2025.03.043. Online ahead of print.

ABSTRACT

OBJECTIVE: The initial screening of laryngeal tumors via voice acoustic analysis is based on the clinician's experience that is subjective. This article introduces a Siamese network with an auxiliary gender classifier for automated, accurate, and objective initial screening of laryngeal tumors based on voice signals.

METHODS: The study involved 71 tumor patients and 293 non-tumor subjects of Chinese Mandarin. This dataset was divided into a training set and a test set in a ratio of 4:1. We applied nine data augmentation techniques to enlarge the voice training set and extracted the corresponding mel-frequency cepstral coefficients (MFCC) maps. The MFCC maps were randomly paired and fed into the proposed Siamese network to achieve multitask classification for tumor and non-tumor, woman and man. The performance of the proposed model was compared with one machine learning method and six classical deep learning models with and without the auxiliary gender classifier.

RESULTS: Experiments demonstrate the superiority of the proposed network compared with the reference models. The proposed model achieved an overall accuracy of 0.9437, an F score of 0.8462, a precision of 0.9167, a sensitivity of 0.7857, and a specificity of 0.9825.

CONCLUSION: The proposed network can assist in the initial screening of laryngeal tumors through voice acoustic analysis. The initial screening solely through voice acoustic analysis can help individuals seek medical assistance outside the hospitals and reduce the burden on doctors as well.

PMID:40318998 | DOI:10.1016/j.jvoice.2025.03.043

Acad Radiol. 2025 May 2:S1076-6332(24)00937-1. doi: 10.1016/j.acra.2024.11.062. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: To explore the feasibility of deep learning (DL)-enhanced, fully automated bone mineral density (BMD) measurement using the ultralow-voltage 80 kV chest CT scans performed for lung cancer screening.

MATERIALS AND METHODS: This study involved 987 patients who underwent 80 kV chest and 120 kV lumbar CT from January to July 2024. Patients were collected from six CT scanners and divided into the training, validation, and test sets 1 and 2 (561: 177: 112: 137). Four convolutional neural networks (CNNs) were employed for automated segmentation (3D VB-Net and SCN), region of interest extraction (3D VB-Net), and BMD calculation (DenseNet and ResNet) of the target vertebrae (T12-L2). The BMD values of T12-L2 were obtained using 80 and 120 kV quantitative CT (QCT), the latter serving as the standard reference. Linear regression and Bland-Altman analyses were used to compare BMD values between 120 kV QCT and 80 kV CNNs, and between 120 kV QCT and 80 kV QCT. Receiver operating characteristic curve analysis was used to assess the diagnostic performance of the 80 kV CNNs and 80 kV QCT for osteoporosis and low BMD from normal BMD.

RESULTS: Linear regression and Bland-ltman analyses revealed a stronger correlation (R2=0.991-0.998 and 0.990-0.991, P<0.001) and better agreement (mean error, -1.36 to 1.62 and 1.72 to 2.27 mg/cm3; 95% limits of agreement, -9.73 to 7.01 and -5.71 to 10.19mg/cm3) for BMD between 120 kV QCT and 80 kV CNNs than between 120 kV QCT and 80 kV QCT. The areas under the curve of the 80 kV CNNs and 80 kV QCT in detecting osteoporosis and low BMD were 0.997-1.000 and 0.997-0.998, and 0.998-1.000 and 0.997, respectively.

CONCLUSION: The DL method could achieve fully automated BMD calculation for opportunistic osteoporosis screening with high accuracy using ultralow-voltage 80 kV chest CT performed for lung cancer screening.

PMID:40318972 | DOI:10.1016/j.acra.2024.11.062

Sex Transm Infect. 2025 May 2:sextrans-2024-056382. doi: 10.1136/sextrans-2024-056382. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aimed to develop robust machine learning (ML)-based and deep learning (DL)-based models capable of detecting mpox cases for surveillance efforts using clinical notes.

METHODS: As part of a learning health system initiative, we conducted a retrospective study of clinical encounters at the Columbia University Irving Medical Center in New York City. We included patients with mpox diagnoses confirmed by PCR testing between 15 May 2022 and 15 October 2022 and three matched controls for each case based on patient age, sex, race, ethnicity and visit month. We trained three mpox surveillance models using: (1) logistic regression with L1 regularisation (least absolute shrinkage and selection operator (LASSO)), (2) ClinicalBERT and (3) ClinicalLongformer. We evaluated model performance using precision, recall, F1 score, area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC) and recall at 80% precision (RP80).

RESULTS: The study included 228 PCR-confirmed mpox cases and 698 controls. LASSO regression outperformed the DL models with a precision, recall and F1 score of 0.93, AUROC of 0.97, AUPRC of 0.93 and RP80 of 0.89. ClinicalBERT achieved a precision of 0.88, recall of 0.89, F1 score of 0.88 and AUROC of 0.93. ClinicalLongformer achieved a precision of 0.87, recall of 0.88, F1 score of 0.87 and AUROC of 0.92. Phrases related to symptoms (eg, lesions and pain) were among the most predictive features in LASSO regression.

CONCLUSIONS: ML and DL models based on clinical notes show promise for identifying mpox cases. In this study, LASSO regression outperformed DL models and excelled in minimising false positives. These findings highlight the potential for ML and DL methods to support case surveillance for mpox and other infectious diseases. These methods may also prove helpful for flagging missed or delayed diagnoses as part of continuous quality improvement.

PMID:40318862 | DOI:10.1136/sextrans-2024-056382

J Adv Res. 2025 May 1:S2090-1232(25)00292-9. doi: 10.1016/j.jare.2025.04.046. Online ahead of print.

ABSTRACT

INTRODUCTION: Activity cliff (AC) is a phenomenon that a pair of similar molecules differ by a small structural alternation but exhibit a large difference in their biochemical activities. This phenomenon affects various tasks ranging from virtual screening to lead optimization in drug development. The AC of small molecules has been extensively investigated but limited knowledge is accumulated about the AC phenomenon in pharmaceutical peptides with canonical amino acids.

OBJECTIVES: This study introduces a quantitative definition and benchmarking framework AMPCliff for the AC phenomenon in antimicrobial peptides (AMPs) composed by canonical amino acids.

METHODS: This study establishes a benchmark dataset of paired AMPs in Staphylococcus aureus from the publicly available AMP dataset GRAMPA, and conducts a rigorous procedure to evaluate various AMP AC prediction models, including nine machine learning, four deep learning algorithms, four masked language models, and four generative language models.

RESULTS: A comprehensive analysis of the existing AMP dataset reveals a significant prevalence of AC within AMPs. AMPCliff quantifies the activities of AMPs by the metric minimum inhibitory concentration (MIC), and defines 0.9 as the minimum threshold for the normalized BLOSUM62 similarity score between a pair of aligned peptides with at least two-fold MIC changes. Our analysis reveals that these models are capable of detecting AMP AC events and the pre-trained protein language model ESM2 demonstrates superior performance across the evaluations. The predictive performance of AMP activity cliffs remains to be further improved, considering that ESM2 with 33 layers only achieves the Spearman correlation coefficient 0.4669 for the regression task of the -log(MIC) values on the benchmark dataset.

CONCLUSION: Our findings highlight limitations in current deep learning-based representation models. To more accurately capture the properties of antimicrobial peptides (AMPs), it is essential to integrate atomic-level dynamic information that reflects their underlying mechanisms of action.

PMID:40318764 | DOI:10.1016/j.jare.2025.04.046

J Hazard Mater. 2025 Apr 26;494:138411. doi: 10.1016/j.jhazmat.2025.138411. Online ahead of print.

ABSTRACT

Oocyte quality is essential for successful pregnancy, yet no discriminant criterion exists to assess the effects of environmental pollutants on oocyte abnormalities. We developed a stepwise framework integrating deep learning-extracted latent features with observable human-concept features focused on toxicity detection, subtype and strength classification. Based on 2126 murine oocyte images, this method achieves performance surpassing human capabilities with ROC-AUC of 0.9087 for toxicity detection, 0.7956-0.9034 for subtype classification with Perfluorohexanesulfonic Acid(PFHxS) achieving highest score of 0.9034 and 0.6434-0.9062 for toxicity strength classification with PFHxS achieving highest score of 0.9062. Notably, Ablation studies confirmed feature fusion improved performance by 18.7-23.4 % over single-domain models, highlighting their complementary relationship. Personalized heatmaps and feature importance revealed biomarker regions such as polar body and cortical areas aligning with clinical knowledge. AI-driven oocyte selection predicts embryo competence under pollutants, bridging computational toxicology to mitigate infertility.

PMID:40318589 | DOI:10.1016/j.jhazmat.2025.138411

Comput Methods Programs Biomed. 2025 Apr 23;267:108803. doi: 10.1016/j.cmpb.2025.108803. Online ahead of print.

ABSTRACT

BACKGROUND: Pathologic complete response (pCR) following neoadjuvant chemotherapy (NACT) is a critical prognostic marker for patients with breast cancer, potentially allowing surgery omission. However, noninvasive and accurate pCR diagnosis remains a significant challenge due to the limitations of current imaging techniques, particularly in cases where tumors completely disappear post-NACT.

METHODS: We developed a novel framework incorporating Dimensional Accumulation for Layered Images (DALI) and an Attention-Box annotation tool to address the unique challenge of analyzing imaging data where target lesions are absent. These methods transform three-dimensional magnetic resonance imaging into two-dimensional representations and ensure consistent target tracking across time-points. Preprocessing techniques, including tissue-region normalization and subtraction imaging, were used to enhance model performance. Imaging features were extracted using radiomics and pretrained deep-learning models, and machine-learning algorithms were integrated into a stacked ensemble model. The approach was developed using the I-SPY 2 dataset and validated with an independent Tangshan People's Hospital cohort.

RESULTS: The stacked ensemble model achieved superior diagnostic performance, with an area under the receiver operating characteristic curve of 0.831 (95 % confidence interval, 0.769-0.887) on the test set, outperforming individual models. Tissue-region normalization and subtraction imaging significantly enhanced diagnostic accuracy. SHAP analysis identified variables that contributed to the model predictions, ensuring model interpretability.

CONCLUSION: This innovative framework addresses challenges of noninvasive pCR diagnosis. Integrating advanced preprocessing techniques improves feature quality and model performance, supporting clinicians in identifying patients who can safely omit surgery. This innovation reduces unnecessary treatments and improves quality of life for patients with breast cancer.

PMID:40318573 | DOI:10.1016/j.cmpb.2025.108803

Phys Med. 2025 May 2;134:104986. doi: 10.1016/j.ejmp.2025.104986. Online ahead of print.

ABSTRACT

BACKGROUND AND PURPOSE: The image quality of single-energy CT (SECT) limited the accuracy of automatic segmentation. Dual-energy CT (DECT) may potentially improve automatic segmentation yet the performance and strategy have not been investigated thoroughly. Based on DECT-generated virtual monochromatic images (VMIs), this study proposed a novel deep learning model (MIAU-Net) and evaluated the segmentation performance on the head organs-at-risk (OARs).

METHODS AND MATERIALS: The VMIs from 40 keV to 190 keV were retrospectively generated at intervals of 10 keV using the DECT of 46 patients. Images with expert delineation were used for training, validation, and testing MIAU-Net for automatic segmentation. Theperformance of MIAU-Net was compared with the existingU-Net, Attention-UNet, nnU-Net and TransFuse methods based on Dice Similarity Coefficient (DSC). Correlationanalysis was performed to evaluate and optimize the impact of different virtual energies on the accuracy of segmentation.

RESULTS: Using MIAU-Net, average DSCs across all virtual energy levels were 93.78 %, 81.75 %, 84.46 %, 92.85 %, 94.40 %, and 84.75 % for the brain stem, optic chiasm, lens, mandible, eyes, and optic nerves, respectively, higher than the previous publications using SECT. MIAU-Net achieved the highest average DSC (88.84 %) and the lowest parameters (14.54 M) in all tested models. The results suggested that 60 keV-80 keV is the optimal VMI energy level for soft tissue delineation, while 100 keV is optimal for skeleton segmentation.

CONCLUSIONS: This work proposed and validated a novel deep learning model for automatic segmentation based on DECT, suggesting potential advantages and OAR-specific optimal energy of using VMIs for automatic delineation.

PMID:40318556 | DOI:10.1016/j.ejmp.2025.104986

Adv Sci (Weinh). 2025 May 3:e2501293. doi: 10.1002/advs.202501293. Online ahead of print.

ABSTRACT

AI-assisted electronic nose systems often emphasize sensitivity-driven datasets, overlooking the comprehensive analysis of gaseous chemical attributes critical for precise gas identification. Conventional fabrication methods generate inconsistent datasets and focus primarily on improving classification accuracy through deep learning, neglecting the fundamental role of sensor material design. This study addresses these challenges by developing a highly reliable sensor platform to standardize gas sensing for deep learning applications. Specifically, 1D SnO2 nanonetworks functionalized with Au and Pd nanocatalysts are fabricated via a systematic deposition process, enhancing gas diffusion and reaction kinetics. Stability improvements through controlled aging process reduce the coefficient of variation to below 5% across seven target gases: acetone, hydrogen, ethanol, carbon monoxide, propane, isoprene, and toluene. The platform exhibits exceptional deep learning performance, achieving over 99.5% classification accuracy using a residual network model, even in high-humidity environments (up to 80% relative humidity) and at parts-per-trillion detection limits. This study highlights the synergy between nanostructure engineering and AI, establishing a robust framework for next-generation bioinspired electronic nose systems with enhanced reliability and analytical capability.

PMID:40318170 | DOI:10.1002/advs.202501293

Med Phys. 2025 May 3. doi: 10.1002/mp.17865. Online ahead of print.

ABSTRACT

BACKGROUND: Accurate delineation of the clinical target volume (CTV) is essential in the radiotherapy treatment of soft tissue sarcomas. However, this process is subject to inter-reader variability due to the need for clinical assessment of risk and extent of potential microscopic spread. This can lead to inconsistencies in treatment planning, potentially impacting treatment outcomes. Most existing automatic CTV delineation methods do not account for this variability and can only generate a single CTV for each case.

PURPOSE: This study aims to develop a deep learning-based technique to generate multiple CTV contours for each case, simulating the inter-reader variability in the clinical practice.

METHODS: We employed a publicly available dataset consisting of fluorodeoxyglucose positron emission tomography (FDG-PET), x-ray computed tomography (CT), and pre-contrast T1-weighted magnetic resonance imaging (MRI) scans from 51 patients with soft tissue sarcoma, along with an independent validation set containing five additional patients. An experienced reader drew a contour of the gross tumor volume (GTV) for each patient based on multi-modality images. Subsequently, two additional readers, together with the first one, were responsible for contouring three CTVs in total based on the GTV. We developed a diffusion model-based deep learning method that is capable of generating arbitrary number of different and plausible CTVs to mimic the inter-reader variability in CTV delineation. The proposed model incorporates a separate encoder to extract features from the GTV masks, leveraging the critical role of GTV information in accurate CTV delineation.

RESULTS: The proposed diffusion model demonstrated superior performance with the highest Dice Index (0.902 compared to values below 0.881 for state-of-the-art models) and the best generalized energy distance (GED) (0.209 compared to values exceeding 0.221 for state-of-the-art models). It also achieved the second-highest recall and precision metrics among the compared ambiguous image segmentation models. Results from both datasets exhibited consistent trends, reinforcing the reliability of our findings. Additionally, ablation studies exploring different model structures and input configurations highlighted the significance of incorporating prior GTV information for accurate CTV delineation.

CONCLUSIONS: The proposed diffusion model successfully generates multiple plausible CTV contours for soft tissue sarcomas, effectively capturing inter-reader variability in CTV delineation.

PMID:40317577 | DOI:10.1002/mp.17865

Med Phys. 2025 May 3. doi: 10.1002/mp.17862. Online ahead of print.

ABSTRACT

BACKGROUND: Patients with early-stage non-small cell lung cancer (NSCLC) typically receive surgery as their primary form of treatment. However, studies have shown that a high proportion of these patients will experience a recurrence after their resection, leading to an increased risk of death. Cancer staging is currently the gold standard for establishing a patient's prognosis and can help clinicians determine patients who may benefit from additional therapy. However, medical images which are used to help determine the cancer stage, have been shown to hold unutilized prognostic information that can augment clinical data and better identify high-risk NSCLC patients. There remains an unmet need for models to incorporate clinical, pathological, surgical, and imaging information, and extend beyond the current staging system to assist clinicians in identifying patients who could benefit from additional therapy immediately after surgery.

PURPOSE: We aimed to determine whether a deep learning model (DLM) integrating FDG PET and CT imaging from the thoracic cavity along with clinical, surgical, and pathological information can predict NSCLC recurrence-free survival (RFS) and stratify patients into risk groups better than conventional staging.

MATERIALS AND METHODS: Surgically resected NSCLC patients enrolled between 2009 and 2018 were retrospectively analyzed from two academic institutions (local institution: 305 patients; external validation: 195 patients). The thoracic cavity (including the lungs, mediastinum, pleural interfaces, and thoracic vertebrae) was delineated on the preoperative FDG PET and CT images and combined with each patient's clinical, surgical, and pathological information. Using the local cohort of patients, a multi-modal DLM using these features was built in a training cohort (n = 225), tuned on a validation cohort (n = 45), and evaluated on testing (n = 35) and external validation (n = 195) cohorts to predict RFS and stratify patients into risk groups. The area under the curve (AUC), Kaplan-Meier curves, and log-rank test were used to assess the prognostic value of the model. The DLM's stratification performance was compared to the conventional staging stratification.

RESULTS: The multi-modal DLM incorporating imaging, pathological, surgical, and clinical data predicted RFS in the testing cohort (AUC = 0.78 [95% CI:0.63-0.94]) and external validation cohort (AUC = 0.66 [95% CI:0.58-0.73]). The DLM significantly stratified patients into high, medium, and low-risk groups of RFS in both the testing and external validation cohorts (multivariable log-rank p < 0.001) and outperformed conventional staging. Conventional staging was unable to stratify patients into three distinct risk groups of RFS (testing: p = 0.94; external validation: p = 0.38). Lastly, the DLM displayed the ability to further stratify patients significantly into sub-risk groups within each stage in the testing (stage I: p = 0.02, stage II: p = 0.03) and external validation (stage I: p = 0.05, stage II: p = 0.03) cohorts.

CONCLUSION: This is the first study to use multi-modality imaging along with clinical, surgical, and pathological data to predict RFS of NSCLC patients after surgery. The multi-modal DLM better stratified patients into risk groups of poor outcomes when compared to conventional staging and further stratified patients within each staging classification. This model has the potential to assist clinicians in better identifying patients that may benefit from additional therapy.

PMID:40317503 | DOI:10.1002/mp.17862

Radiol Med. 2025 May 3. doi: 10.1007/s11547-025-02013-y. Online ahead of print.

ABSTRACT

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related deaths worldwide. Intermediate-stage HCC is often treated with either transcatheter arterial chemoembolisation (TACE) or transcatheter arterial embolisation (TAE). Integrating machine learning (ML) offers the possibility of improving treatment outcomes through enhanced patient selection. This systematic review evaluates the effectiveness of ML models in improving the precision and efficacy of both TACE and TAE for intermediate-stage HCC. A comprehensive search of PubMed, EMBASE, Web of Science, and Cochrane Library databases was conducted for studies applying ML models to TACE and TAE in patients with intermediate-stage HCC. Seven studies involving 4,017 patients were included. All included studies were from China. Various ML models, including deep learning and radiomics, were used to predict treatment response, yielding a high predictive accuracy (AUC 0.90). However, study heterogeneity limited comparisons. While ML shows potential in predicting treatment outcomes, further research with standardised protocols and larger, multi-centre trials is needed for clinical integration.

PMID:40317437 | DOI:10.1007/s11547-025-02013-y