ACM Trans Intell Syst Technol. 2025 Jun;16(3):71. doi: 10.1145/3718097. Epub 2025 Jun 10.

ABSTRACT

Precise estimation of treatment effects is crucial for accurately evaluating the intervention. While deep learning models have exhibited promising performance in learning counterfactual representations for treatment effect estimation (TEE), a major limitation in most of these models is that they often overlook the diversity of treatment effects across potential subgroups that have varying treatment effects and characteristics, treating the entire population as a homogeneous group. This limitation restricts the ability to precisely estimate treatment effects and provide targeted treatment recommendations. In this paper, we propose a novel treatment effect estimation model, named SubgroupTE, which incorporates subgroup identification in TEE. SubgroupTE identifies heterogeneous subgroups with different responses and more precisely estimates treatment effects by considering subgroup-specific treatment effects in the estimation process. In addition, we introduce an expectation-maximization (EM)-based training process that iteratively optimizes estimation and subgrouping networks to improve both estimation and subgroup identification. Comprehensive experiments on the synthetic and semi-synthetic datasets demonstrate the outstanding performance of SubgroupTE compared to the existing works for treatment effect estimation and subgrouping models. Additionally, a real-world study demonstrates the capabilities of SubgroupTE in enhancing targeted treatment recommendations for patients with opioid use disorder (OUD) by incorporating subgroup identification with treatment effect estimation.

PMID:40575765 | PMC:PMC12199269 | DOI:10.1145/3718097

Appl Plant Sci. 2025 Jun 7;13(3):e70010. doi: 10.1002/aps3.70010. eCollection 2025 May-Jun.

ABSTRACT

PREMISE: This study investigates advanced training techniques to improve the performance of convolutional neural networks for disease detection in cocoa, Theobroma cacao.

METHODS: Despite recent stagnation in accuracy improvements in computer vision for image classification, our research demonstrates significant advancements in performance through semi-supervised learning, specialised loss functions, and the inclusion of a non-cocoa class.

RESULTS: Semi-supervised learning reduced overfitting and enhanced generalisability, particularly for subtle symptoms. The non-cocoa class exposed models to a broad range of relevant features, significantly improving model robustness and performance in difficult cases. Grad-CAM for qualitative assessment provided valuable insights into model behaviour, highlighting cases of overfitting missed by summary statistics. We also describe dynamic focal loss, a novel loss function that uses an empirical measure of difficulty to weight each image. Our results suggest that while PhytNet shows promise in terms of computational efficiency and superior handling of difficult images, ResNet18 with semi-supervised learning and dynamic focal loss emerged as the strongest contender for real-world deployment.

DISCUSSION: This research underscores the potential of semi-supervised learning and advanced loss functions in enhancing the applicability of deep learning models in agricultural disease management. It also presents a new high-quality benchmark dataset of 7220 images of diseased and healthy cocoa trees, offering a much greater and more realistic challenge than the Plan Village dataset.

PMID:40575549 | PMC:PMC12188622 | DOI:10.1002/aps3.70010

PCN Rep. 2025 Jun 26;4(2):e70134. doi: 10.1002/pcn5.70134. eCollection 2025 Jun.

ABSTRACT

BACKGROUND: Two of the most common complaints seen in neurology clinics are Alzheimer's disease (AD) and mild cognitive impairment (MCI), characterized by similar symptoms. The aim of this study was to develop and internally validate the diagnostic value of combined neurological and radiological predictors in differentiating mild AD from MCI as the outcome variable, which helps in preventing AD development.

METHODS: A cross-sectional study of 161 participants was conducted in a general healthcare setting, including 30 controls, 71 mild AD, and 60 MCI. Binary logistic regression was used to identify predictors of interest, with collinearity assessment conducted prior to model development. Model performance was assessed through calibration, shrinkage, and decision-curve analyses. Finally, the combined clinical and radiological model was compared to models utilizing only clinical or radiological predictors.

RESULTS: The final model included age, sex, education status, Montreal cognitive assessment, Global Cerebral Atrophy Index, Medial Temporal Atrophy Scale, mean hippocampal volume, and Posterior Parietal Atrophy Index, with the area under the curve of 0.978 (0.934-0.996). Internal validation methods did not show substantial reduction in diagnostic performance. Combined model showed higher diagnostic performance compared to clinical and radiological models alone. Decision curve analysis highlighted the usefulness of this model for differentiation across all probability levels.

CONCLUSION: A combined clinical-radiological model has excellent diagnostic performance in differentiating mild AD from MCI. Notably, the model leveraged straightforward neuroimaging markers, which are relatively simple to measure and interpret, suggesting that they could be integrated into practical, formula-driven diagnostic workflows without requiring computationally intensive deep learning models.

PMID:40575445 | PMC:PMC12199059 | DOI:10.1002/pcn5.70134

Front Endocrinol (Lausanne). 2025 Jun 12;16:1556681. doi: 10.3389/fendo.2025.1556681. eCollection 2025.

ABSTRACT

OBJECTIVE: To evaluate the predictive performance of a convolutional neural network for analyzing electronic medical records in assisted reproductive therapy and to compare its accuracy and interpretability with traditional machine learning models. The study also explores the feasibility of deploying such models in resource-limited clinical settings.

DESIGN: Retrospective cohort study based on EMR data using five models: CNN, Naïve Bayes, Random Forest, Decision Tree, and Feedforward Neural Network. Feature importance and model interpretability were evaluated using SHAP.

SETTING: First Hospital of Zhengzhou University.

POPULATION: 48,514 fresh IVF cycles from August 2009 to May 2018.

METHODS: Preprocessed EMR data were used to train and evaluate five classification models predicting live birth outcomes. Stratified 5-fold cross-validation was performed for robust performance estimation. ROC curves and AUC values were used for comparative evaluation.

MAIN OUTCOME MEASURE: Live birth.

RESULTS: The CNN model achieved an accuracy of 0.9394 ± 0.0013, AUC of 0.8899 ± 0.0032, precision of 0.9348 ± 0.0018, recall of 0.9993 ± 0.0012, and F1 score of 0.9660 ± 0.0007. Its performance was comparable to Random Forest (accuracy: 0.9406 ± 0.0017, AUC: 0.9734 ± 0.0012), and superior to Decision Tree, Naïve Bayes, and Feedforward Neural Network in recall and robustness. CNN demonstrated stable convergence during training, and SHAP-based interpretation highlighted maternal age, BMI, antral follicle count, and gonadotropin dosage as the top predictors for live birth outcome.

CONCLUSIONS: With appropriate input transformation, CNNs can effectively model structured EMR data and offer predictive performance comparable to ensemble methods. Their scalability, high sensitivity, and interpretability make CNNs promising candidates for integration into clinical workflows, particularly in environments with limited computational resources.

PMID:40575261 | PMC:PMC12197960 | DOI:10.3389/fendo.2025.1556681

Front Oncol. 2025 Jun 12;15:1581157. doi: 10.3389/fonc.2025.1581157. eCollection 2025.

ABSTRACT

Ovarian cancer, as a common gynecological malignancy, is often found at an advanced stage clinically. Thus, improving the early diagnosis of ovarian cancer is crucial for the survival rate of patients. Ultrasound examination is the main method for ovarian cancer screening, but it is greatly influenced by the operator's experience and technique, increasing the risk of misdiagnosis and missed diagnosis. Artificial intelligence uses computers to learn from input data and has already made significant progress in image recognition. Applying artificial intelligence to ultrasound diagnosis of ovarian cancer can enhance diagnostic accuracy, providing earlier treatment for patients. This article reviews the current application of artificial intelligence in the ultrasound diagnosis of ovarian cancer, in order to provide a reference for subsequent clinical diagnosis and treatment.

PMID:40575169 | PMC:PMC12198115 | DOI:10.3389/fonc.2025.1581157

Patterns (N Y). 2025 Apr 25;6(6):101239. doi: 10.1016/j.patter.2025.101239. eCollection 2025 Jun 13.

ABSTRACT

Existing antibody language models (AbLMs) are pre-trained using a masked language modeling (MLM) objective with uniform masking probabilities. While these models excel at predicting germline residues, they often struggle with mutated and non-templated residues, which concentrate in the complementarity-determining regions (CDRs) and are crucial for antigen binding specificity. Here, we demonstrate that preferential masking of the primarily non-templated CDR3 is a compute-efficient strategy to enhance model performance. We pre-trained two AbLMs using either uniform or preferential masking and observed that the latter improves residue prediction accuracy in the highly variable CDR3. Preferential masking also improves antibody classification by native chain pairing and binding specificity, suggesting improved CDR3 understanding and indicating that non-random, learnable patterns help govern antibody chain pairing. We further show that specificity classification is largely informed by residues in the CDRs, demonstrating that AbLMs learn meaningful patterns that align with immunological understanding.

PMID:40575131 | PMC:PMC12191730 | DOI:10.1016/j.patter.2025.101239

Patterns (N Y). 2025 May 30;6(6):101285. doi: 10.1016/j.patter.2025.101285. eCollection 2025 Jun 13.

ABSTRACT

In liquid biopsy, detecting and differentiating circulating tumor cells (CTCs) and non-CTCs in metastatic cancer patients' blood samples remains challenging. The current gold standard often involves tedious manual examination of extensive image galleries. While machine learning (ML) offers potential automation, human expertise remains essential, particularly when ML systems face uncertainty or incorrect predictions due to limited labeled data. Combining self-supervised deep learning with an easily adaptable conventional ML classifier, we propose a human-in-the-loop approach with a targeted sampling strategy. By directing human efforts to label a limited set of new training samples from high-uncertainty clusters in the latent space, we iteratively reduce the system's uncertainty and improve classification performance, thereby saving time compared to naive sampling approaches. On data from metastatic breast cancer patients, we show the feasibility of our approach and achieve better performance while reducing expert evaluation time compared to the gold standard, the FDA-approved CellSearch system.

PMID:40575126 | PMC:PMC12191740 | DOI:10.1016/j.patter.2025.101285

Patterns (N Y). 2025 May 6;6(6):101262. doi: 10.1016/j.patter.2025.101262. eCollection 2025 Jun 13.

ABSTRACT

We describe NLSExplorer, an interpretable approach for nuclear localization signal (NLS) prediction. By utilizing the extracted information on nuclear-specific sites from the protein language model to assist in NLS detection, NLSExplorer achieves superior performance with greater than 10% improvement in the F1 score compared with existing methods on benchmark datasets and highlights other nuclear transport segments. We applied NLSExplorer to the nucleus-localized proteins in the Swiss-Prot database to extract valuable segments. A comprehensive analysis of these segments revealed a potential NLS landscape and uncovered features of nuclear transport segments across 416 species. This study introduces a powerful tool for exploring the NLS universe and provides a versatile network that can efficiently detect characteristic domains and motifs.

PMID:40575124 | PMC:PMC12191761 | DOI:10.1016/j.patter.2025.101262

Zhongguo Yi Liao Qi Xie Za Zhi. 2025 May 30;49(3):250-254. doi: 10.12455/j.issn.1671-7104.240561.

ABSTRACT

Medical equipment management plays a crucial role in enhancing the quality and efficiency of healthcare services. However, traditional management approaches are increasingly inadequate to meet the growing demands of modern healthcare. As intelligent operation and maintenance (O&M) models based on big data, the Internet of Things (IoT), and artificial intelligence (AI) technologies develop, it is imperative to explore their application in medical equipment management. This paper reviews the technical overview of intelligent O&M and discusses the algorithms and challenges of intelligent O&M models based on different technologies. It also proposes issues that need improvement in intelligent O&M models, aiming to provide valuable references for the future development of medical equipment management.

PMID:40574433 | DOI:10.12455/j.issn.1671-7104.240561

Zhongguo Yi Liao Qi Xie Za Zhi. 2025 May 30;49(3):237-243. doi: 10.12455/j.issn.1671-7104.240499.

ABSTRACT

Computer-assisted methods for pathological image analysis can improve doctor's efficiency of image reading and diagnostic accuracy, effectively addressing the shortage of pathology diagnostic manpower. With the rapid development of artificial intelligence and digital pathology, deep learning technology has spurred a wealth of research in the field of histopathology. This article reviews the various applications of deep learning in digital pathological image analysis, such as pathological image segmentation, cancer auxiliary diagnosis, and cancer prognosis prediction, and discusses the challenges and solutions in its application. Furthermore, it predicts future trends in deep learning for pathological image analysis and proposes potential research directions.

PMID:40574431 | DOI:10.12455/j.issn.1671-7104.240499

Cell Rep Med. 2025 Jun 19:102203. doi: 10.1016/j.xcrm.2025.102203. Online ahead of print.

ABSTRACT

Systemic lupus erythematosus (SLE) is a serious autoimmune disorder predominantly affecting women. However, screening for SLE and related complications poses significant challenges globally, due to complex diagnostic criteria and public unawareness. Since SLE-related retinal involvement could provide insights into disease activity and severity, we develop a deep learning system (DeepSLE) to detect SLE and its retinal and kidney complications from retinal images. In multi-ethnic validation datasets comprising 247,718 images from China and UK, DeepSLE achieves areas under the receiver operating characteristic curve of 0.822-0.969 for SLE. Additionally, DeepSLE demonstrates robust performance across subgroups stratified by gender, age, ethnicity, and socioeconomic status. To ensure DeepSLE's explainability, we conduct both qualitative and quantitative analyses. Furthermore, in a prospective reader study, DeepSLE demonstrates higher sensitivities compared with primary care physicians. Altogether, DeepSLE offers digital solutions for detecting SLE and related complications from retinal images, holding potential for future clinical deployment.

PMID:40570853 | DOI:10.1016/j.xcrm.2025.102203

Comput Biol Med. 2025 Jun 25;195:110592. doi: 10.1016/j.compbiomed.2025.110592. Online ahead of print.

ABSTRACT

Colorectal cancer (CRC) is a leading cause of death worldwide. A plethora of research shows the alteration of the gut microbiome and the association of bacterial taxa with CRC. Gaining insights into the health status through microbiome-based diagnosis is a rapidly growing area of research. Many studies have utilized machine learning (ML) to leverage gut microbial dysbiosis for CRC screening, yet most have been limited by their training data and algorithms. Here, using 1728 publicly available metagenomic samples from 11 studies across eight countries, we developed a web-based tool, "CRCpred," employing ML and deep learning-based hybrid algorithms for CRC prediction. The XGBoost algorithm demonstrated the highest performance, achieving an average area under the curve (AUC) of 0.90 on the test and 0.91 on the validation datasets. Our results highlight the utility of CRCpred in predicting CRC and healthy status using gut bacterial species relative abundance profile. CRCpred is publicly available at https://metabiosys.iiserb.ac.in/crcpred.

PMID:40570762 | DOI:10.1016/j.compbiomed.2025.110592

Environ Monit Assess. 2025 Jun 27;197(7):814. doi: 10.1007/s10661-025-14261-6.

ABSTRACT

Coral reef managers require various forms of data. While monitoring is typically the preserve of scientists, there is an increasing need to collect larger scale, up-to-date data to prioritise limited conservation resources. Citizen science combined with novel technology may achieve data collection at the required scale, but the accuracy and feasibility of new tools must be assessed. Here, we show that a citizen science program that collects large field-of-view benthic images and analyses them using a combination of deep learning and online citizen scientists can produce accurate benthic cover estimates of key coral groups. The deep learning and citizen scientist analysis methods had different but complementary strengths depending on coral category. When the best performing analysis method was used for each category in all images, mean estimates from 8086 images of percent benthic cover of branching Acropora, plating Acropora, and massive-form coral were ~ 99% accurate compared to expert assessment, and > 95% accurate at all coral cover ranges tested. Site-level accuracy of 95% was attainable with 18-80 images. Power analyses showed that up to 114 images per site were needed to detect a 10% absolute difference in coral cover per category (power = 0.8). However, estimates of 'all other coral' as a single category achieved 95% accuracy at only 60% of sites and for images with 10-30% coral cover. Overall, emerging technology and citizen science present an attainable tool for collecting inexpensive, widespread data that can complement higher resolution survey programs or be an accessible tool for locations with limited scientific or conservation resources.

PMID:40571887 | DOI:10.1007/s10661-025-14261-6

Insights Imaging. 2025 Jun 26;16(1):135. doi: 10.1186/s13244-025-02011-8.

ABSTRACT

OBJECTIVES: Post-surgical prediction of recurrence or metastasis for primary gastrointestinal stromal tumors (GISTs) remains challenging. We aim to develop individualized clinical follow-up strategies for primary GIST patients, such as shortening follow-up time or extending drug administration based on the clinical deep learning radiomics model (CDLRM).

METHODS: The clinical information on primary GISTs was collected from two independent centers. Postoperative recurrence or metastasis in GIST patients was defined as the endpoint of the study. A total of nine machine learning models were established based on the selected features. The performance of the models was assessed by calculating the area under the curve (AUC). The CDLRM with the best predictive performance was constructed. Decision curve analysis (DCA) and calibration curves were analyzed separately. Ultimately, our model was applied to the high-potential malignant group vs the low-malignant-potential group. The optimal clinical application scenarios of the model were further explored by comparing the DCA performance of the two subgroups.

RESULTS: A total of 526 patients, 260 men and 266 women, with a mean age of 62 years, were enrolled in the study. CDLRM performed excellently with AUC values of 0.999, 0.963, and 0.995 for the training, external validation, and aggregated sets, respectively. The calibration curve indicated that CDLRM was in good agreement between predicted and observed probabilities in the validation cohort. The results of DCA's performance in different subgroups show that it was more clinically valuable in populations with high malignant potential.

CONCLUSION: CDLRM could help the development of personalized treatment and improved follow-up of patients with a high probability of recurrence or metastasis in the future.

CRITICAL RELEVANCE STATEMENT: This model utilizes imaging features extracted from CT scans (including radiomic features and deep features) and clinical data to accurately predict postoperative recurrence and metastasis in patients with primary GISTs, which has a certain auxiliary role in clinical decision-making.

KEY POINTS: We developed and validated a model to predict recurrence or metastasis in patients taking oral imatinib after GIST. We demonstrate that CT image features were associated with recurrence or metastases. The model had good predictive performance and clinical benefit.

PMID:40571854 | DOI:10.1186/s13244-025-02011-8

Sci Rep. 2025 Jun 26;15(1):20313. doi: 10.1038/s41598-025-03560-0.

ABSTRACT

Arabic sign language (ArSL) is a visual-manual language which facilitates communication among Deaf people in the Arabic-speaking nations. Recognizing the ArSL is crucial due to variety of reasons, including its impact on the Deaf populace, education, healthcare, and society, as well. Previous approaches for the recognition of Arabic sign language have some limitations especially in terms of accuracy and their capability to capture the detailed features of the signs. To overcome these challenges, a new model is proposed namely DeepArabianSignNet, that incorporates DenseNet, EfficientNet and an attention-based Deep ResNet. This model uses a newly introduced G-TverskyUNet3+ to detect regions of interest in preprocessed Arabic sign language images. In addition, employing a novel metaheuristic algorithm, the Crisscross Seed Forest Optimization Algorithm, which combines the Crisscross Optimization and Forest Optimization algorithms to determine the best features from the extracted texture, color, and deep learning features. The proposed model is assessed using two databases, the variation of the training rate was 70% and 80%; Database 2 was exceptional, with an accuracy of 0.97675 for 70% of the training data and 0.98376 for 80%. The results presented in this paper prove that DeepArabianSignNet is effective in improving Arabic sign language recognition.

PMID:40571695 | DOI:10.1038/s41598-025-03560-0

J Mol Biol. 2025 Jun 24:169305. doi: 10.1016/j.jmb.2025.169305. Online ahead of print.

ABSTRACT

Allosteric regulation in proteins is a critical aspect of cellular function, influencing various biological processes through conformational or dynamic changes induced by effector molecules. Allosteric drugs possess significant therapeutic value due to their unique advantages, such as high specificity and diverse regulatory types, yet their presence in clinical applications remains limited. Understanding the relationship between protein sequence, structure, and allosteric regulation can promote insights into allosteric mechanisms and facilitate allosteric drug design. In this review, we present an overview of marketed allosteric drugs, summarize recent computational methods for predicting allosteric sites based on protein sequences and structures, together with case studies of recent rational allosteric drug design. We also discuss challenges and future directions in computer-aided allosteric drug design, with emphasis on the potential of multi-modal data integration and interpretable deep learning models in improving allosteric site prediction and rational allosteric drug design.

PMID:40571274 | DOI:10.1016/j.jmb.2025.169305

J Invest Dermatol. 2025 Jun 24:S0022-202X(25)02144-X. doi: 10.1016/j.jid.2025.06.1567. Online ahead of print.

ABSTRACT

Mycosis fungoides (MF) is characterized by widespread skin patches that may progress to plaques and tumors, necessitating precise tumor burden assessment for staging and treatment guidance. However, existing methods, including the widely accepted modified Severity Weighted Assessment Tool (mSWAT), present significant challenges in routine practice due to their time-consuming nature and inter-observer variability. This study developed an AI model, mSWAT-Net, to estimate mSWAT scores using clinical images of MF patients. Notably, the overlap area segmentation sub-module of mSWAT-Net addressed double-counting errors in multi-angle photos through training on 3,904 annotated images generated from 61 three-dimensional (3D) human images. Across 2,463 standardized full-body photographs from 134 imaging series, mSWAT-Net demonstrated comparable performance to experienced cutaneous lymphoma specialists, achieving intraclass correlation coefficients (ICCs) of 0.917 (internal validation) and 0.846 (temporal validation) for mSWAT score. Moreover, mSWAT-Net outperformed three junior dermatologists in image-based scoring (ICC: 0.917 vs. 0.777), and demonstrated robust performance when compared to ground truth derived from 3D patient imaging (ICC: 0.812). Finally, mSWAT-Net was deployed as a free web application to support MF management in clinical settings. These findings highlight the potential of mSWAT-Net as an accurate, automated clinical tool for facilitating patient follow-up, treatment monitoring, and remote consultations.

PMID:40571158 | DOI:10.1016/j.jid.2025.06.1567

J Microbiol Methods. 2025 Jun 24:107185. doi: 10.1016/j.mimet.2025.107185. Online ahead of print.

ABSTRACT

Colony counting plays a crucial role in evaluating food quality and safety. The segmentation of colony adhesion images can significantly enhance the accuracy of food safety assessments. To achieve high-precision segmentation of colony adhesion images, this paper presents a novel multi-scale feature deep-enhanced fusion network MFDF-UNet, specifically designed for colony adhesion image segmentation. The core of the network lies in the design of a self-similar fusion fractal structure, which recursively integrates layers to enhance the network's ability to extract, integrate, and transfer multi-scale feature information. The DEC (depth-enhanced connectivity) units and PF (progressive fusion) modules in each stage progressively accumulate detailed features, thus improving the network's capacity to handle complex structures. Additionally, the design strengthens the information transfer between different layers, ensuring consistency of features across multiple layers. This reduces the imbalance in feature information transfer that can occur when certain regions of the image contain prominent edges, textures, or structural features, while other areas are relatively blurred or lack distinct features.The MFDF-UNet model achieved an average segmentation accuracy of 77.95 %, precision of 97.55 %, and a mean intersection-over-union (mIoU) of 57.94 % on the AGAR-based hybrid colony adhesion segmentation test dataset. Compared to other deep learning methods, such as PSPNet, DeepLabv3+, SegFormer, YOLOv8, U-Net, and ResNet, MFDF-UNet outperforms the highest-performing ResUNet by 7.53 % in segmentation accuracy, improves precision by 1.5 %, and surpasses ResUNet by 4.82 % in mIoU.Although our model requires slightly more parameters and training time, the improvements in segmentation accuracy and image quality sufficiently justify the additional cost, demonstrating its potential for practical applications in colony adhesion segmentation.

PMID:40570932 | DOI:10.1016/j.mimet.2025.107185

Phys Med Biol. 2025 Jun 26. doi: 10.1088/1361-6560/ade8ce. Online ahead of print.

ABSTRACT

Beam orientation optimization (BOO) in intensity-modulated radiation therapy (IMRT) is a complex, non-convex problem traditionally addressed with heuristic methods.
Approach: This work demonstrates the potential improvement of the proposed BOO, providing a mathematically grounded benchmark that can guide and validate heuristic BOO methods, while also offering a computationally efficient workflow suitable for clinical application. A novel framework integrating second-order cone programming (SOCP) relaxation, sparse mixed integer programming (SMIP), and deep inverse optimization is proposed. Nonconvex dose-volume constraints were managed via SOCP relaxation, ensuring convexity while maintaining sparsity. BOO was formulated as an SMIP problem with binary beam selection, solved using an augmented Lagrange method. To accelerate optimization, a neural network approximated optimal solution, improving computational efficiency eightfold. A retrospective analysis of 12 locally advanced non-small cell lung cancer (NSCLC) patients (60 Gy prescription) compared automated BOO-selected beam angles with expert selections, evaluating dosimetric metrics such as planning target volume (PTV) maximum dose, D98%, lung V20, and mean heart and esophagus dose.
Main results: In 12 retrospective study, the automated BOO demonstrated superior dose conformity and sparing of critical structures. Specifically, the BOO plans achieved comparable PTV coverage (maximum: 61.7±1.4Gy vs. 62.1±1.5Gy, D98%: 59.5±0.7Gy vs. 59.5±0.6Gy, D2%: 61.2±1.3Gy vs. 61.4±1.4Gy) but demonstrated improved sparing for lungs (V20: 9.8±2.2% vs. 11.5±2.3%), heart (mean: 3.3±0.6Gy vs. 4.3±0.5Gy), esophagus (mean: 0.5±1.3Gy vs. 1.8±1.5Gy), and spinal cord (max: 7.2±3.4Gy vs. 9.0±3.2Gy) compared to human-selected plans. 
Significance: This approach highlighted the potential of BOO to enhance treatment efficacy by optimizing beam angles more effectively than manual selection. This framework establishes a benchmark for BOO in IMRT, enhancing heuristic methods through a hybrid framework that combines mathematical optimization with targeted heuristics to improve solution quality and computational efficiency. The integration of SMIP and deep inverse optimization significantly improves computational efficiency and plan quality.

PMID:40570902 | DOI:10.1088/1361-6560/ade8ce

Phys Med Biol. 2025 Jun 26. doi: 10.1088/1361-6560/ade8cc. Online ahead of print.

ABSTRACT

Low-dose computed tomography (LDCT) denoising plays an important role in medical imaging for reducing the radiation dose to patients. Recently, various data-driven and diffusion-based deep learning (DL) methods have been developed and shown promising results in LDCT denoising. However, challenges remain in ensuring generalizability to different datasets and mitigating uncertainty from stochastic sampling. In this paper, we introduce a novel dose-aware diffusion model that effectively reduces CT image noise while maintaining structural fidelity and being generalizable to different dose levels.
Approach: Our approach employs a physics-based forward process with continuous timesteps, enabling flexible representation of diverse noise levels. We incorporate a computationally efficient noise calibration module in our diffusion framework that resolves misalignment between intermediate results and their corresponding timesteps. Furthermore, we present a simple yet effective method for estimating appropriate timesteps for unseen LDCT images, allowing generalization to an unknown, arbitrary dose levels.
Main Results: Both qualitative and quantitative evaluation results on Mayo Clinic datasets show that the proposed method outperforms existing denoising methods in preserving the noise texture and restoring anatomical structures. The proposed method also shows consistent results on different dose levels and an unseen dataset.
Significance: We propose a novel dose-aware diffusion model for LDCT denoising, aiming to address the generalization and uncertainty issues of existing diffusion-based DL methods. Our experimental results demonstrate the effectiveness of the proposed method across different dose levels. We expect that our approach can provide a clinically practical solution for LDCT denoising with its high structural fidelity and computational efficiency.

PMID:40570896 | DOI:10.1088/1361-6560/ade8cc