Radiol Imaging Cancer. 2025 Mar;7(2):e240161. doi: 10.1148/rycan.240161.

ABSTRACT

Purpose To develop a radiomic graph framework, RadGraph, for spatial analysis of pretreatment CT images to improve prediction of local-regional recurrence (LR) and distant metastasis (DM) in head and neck squamous cell carcinoma (HNSCC). Materials and Methods This retrospective study included four public pre-radiotherapy treatment CT datasets of patients with HNSCC obtained from The Cancer Imaging Archive (images collected between 2003 and 2018). Computational graphs and graph attention deep learning methods were leveraged to holistically model multiple regions in the head and neck anatomy. Clinical features, including age, sex, and human papillomavirus infection status, were collected for a baseline model. Model performance in predicting LR and DM was evaluated via area under the receiver operating characteristic curve (AUC) and qualitative interpretation of model attention. Results A total of 3434 patients (61 years ± 11 [SD], 2774 male) were divided into training (n = 1576), validation (n = 379), and testing (n = 1479) datasets. RadGraph achieved AUCs of up to 0.83 and 0.90 for LR and DM prediction, respectively. RadGraph showed higher performance compared with the clinical baseline (AUCs up to 0.73 for LR prediction and 0.83 for DM prediction) and previously published approaches (AUCs up to 0.81 for LR prediction and 0.87 for DM prediction). Graph attention atlases enabled visualization of regions coinciding with cervical lymph node chains as important for outcome prediction. Conclusion RadGraph leveraged information from tumor and nontumor regions to effectively predict LR and DM in a large multi-institutional dataset of patients with radiation therapy-treated HNSCC. Graph attention atlases enabled interpretation of model predictions. Keywords: CT, Informatics, Neural Networks, Radiation Therapy, Head/Neck, Computer Applications-General (Informatics), Tumor Response, Head and Neck Squamous Cell Carcinoma, Locoregional Recurrence, Radiotherapy, Deep Learning, Radiomics Supplemental material is available for this article. © RSNA, 2025.

PMID:39982207 | DOI:10.1148/rycan.240161

Cytometry A. 2025 Feb 21. doi: 10.1002/cyto.a.24917. Online ahead of print.

ABSTRACT

Alterations in nuclear morphology are useful adjuncts and even diagnostic tools used by pathologists in the diagnosis and grading of many tumors, particularly malignant tumors. Large datasets such as TCGA and the Human Protein Atlas, in combination with emerging machine learning and statistical modeling methods, such as feature extraction and deep learning techniques, can be used to extract meaningful knowledge from images of nuclei, particularly from cancerous tumors. Here, we describe a new technique based on the mathematics of optimal transport for modeling the information content related to nuclear chromatin structure directly from imaging data. In contrast to other techniques, our method represents the entire information content of each nucleus relative to a template nucleus using a transport-based morphometry (TBM) framework. We demonstrate that the model is robust to different staining patterns and imaging protocols, and can be used to discover meaningful and interpretable information within and across datasets and cancer types. In particular, we demonstrate morphological differences capable of distinguishing nuclear features along the spectrum from benign to malignant categories of tumors across different cancer tissue types, including tumors derived from liver parenchyma, thyroid gland, lung mesothelium, and skin epithelium. We believe these proof-of-concept calculations demonstrate that the TBM framework can provide the quantitative measurements necessary for performing meaningful comparisons across a wide range of datasets and cancer types that can potentially enable numerous cancer studies, technologies, and clinical applications and help elevate the role of nuclear morphometry into a more quantitative science.

PMID:39982036 | DOI:10.1002/cyto.a.24917

J Comput Chem. 2025 Mar 5;46(6):e70060. doi: 10.1002/jcc.70060.

ABSTRACT

Scoring functions (SFs) of molecular docking is a vital component of structure-based virtual screening (SBVS). Traditional SFs yield their inherent shortage for idealized approximations and simplifications predicting the binding affinity. Complementarily, SFs based on deep learning (DL) have emerged as powerful tools for capturing intricate features within protein-ligand (PL) interactions. We here present a docking-score fusion strategy that integrates pose scores derived from GNINA's convolutional neural network (CNN) with traditional docking scores. Extensive validation on diverse datasets has shown that by means of multiplying Watvina docking score by CNNscore demonstrates state-of-the-art screening power. Furthermore, in a reverse practice, our docking-score fusion technique was incorporated into the virtual screening (VS) workflow aimed at identifying inhibitors of the challenging target TYK2. Two promising hits with IC50 9.99 μM and 13.76 μM in vitro were identified from nearly 12 billion molecules.

PMID:39981784 | DOI:10.1002/jcc.70060

Phys Imaging Radiat Oncol. 2025 Jan 30;33:100716. doi: 10.1016/j.phro.2025.100716. eCollection 2025 Jan.

ABSTRACT

For 18 months following clinical introduction of deep-learning auto-segmentation (DLAS), an audit of organ at risk (OAR) contour editing was performed, including 1255 patients from a single institution and the majority of tumour sites. Mean surface-Dice similarity coefficient increased from 0.87 to 0.97, the number of unedited OARs increased from 21.5 % to 40 %. The audit identified changes in editing corresponding to vendor model changes, adaption of local contouring practice and reduced editing in areas of no clinical significance. The audit allowed assessment of the level and frequency of editing and identification of outlier cases.

PMID:39981522 | PMC:PMC11840498 | DOI:10.1016/j.phro.2025.100716

Cancer Innov. 2025 Feb 20;4(2):e159. doi: 10.1002/cai2.159. eCollection 2025 Apr.

ABSTRACT

Breast cancer (BC) remains a significant threat to women's health worldwide. The oncology field had an exponential growth in the abundance of medical images, clinical information, and genomic data. With its continuous advancement and refinement, artificial intelligence (AI) has demonstrated exceptional capabilities in processing intricate multidimensional BC-related data. AI has proven advantageous in various facets of BC management, encompassing efficient screening and diagnosis, precise prognosis assessment, and personalized treatment planning. However, the implementation of AI into precision medicine and clinical practice presents ongoing challenges that necessitate enhanced regulation, transparency, fairness, and integration of multiple clinical pathways. In this review, we provide a comprehensive overview of the current research related to AI in BC, highlighting its extensive applications throughout the whole BC cycle management and its potential for innovative impact. Furthermore, this article emphasizes the significance of constructing patient-oriented AI algorithms. Additionally, we explore the opportunities and potential research directions within this burgeoning field.

PMID:39981497 | PMC:PMC11840326 | DOI:10.1002/cai2.159

Front Neurosci. 2025 Feb 5;19:1497881. doi: 10.3389/fnins.2025.1497881. eCollection 2025.

ABSTRACT

This study investigated the impact of brain atlas selection on the classification accuracy of Autism Spectrum Disorder (ASD) models using functional Magnetic Resonance Imaging (fMRI) data. Brain atlases, such as AAL, CC200, Harvard-Oxford, and Yeo 7/17, are used to define regions of interest (ROIs) for fMRI analysis and play a crucial role in enabling researchers to study connectivity patterns and neural dynamics in ASD patients. Through a systematic review, we examined the performance of different atlases in various machine-learning and deep-learning frameworks for ASD classification. The results reveal that atlas selection significantly affects classification accuracy, with denser atlases, such as CC400, providing higher granularity, whereas coarser atlases such as AAL, offer computational efficiency. Furthermore, we discuss the dynamics of combining multiple atlases to enhance feature extraction and explore the implications of atlas selection across diverse datasets. Our findings emphasize the need for standardized approaches to atlas selection and highlight future research directions, including the integration of novel atlases, advanced data augmentation techniques, and end-to-end deep-learning models. This study provides valuable insights into optimizing fMRI-based ASD diagnosis and underscores the importance of interpreting atlas-specific features for an improved understanding of brain connectivity in ASD.

PMID:39981402 | PMC:PMC11841380 | DOI:10.3389/fnins.2025.1497881

Heliyon. 2025 Feb 1;11(3):e42404. doi: 10.1016/j.heliyon.2025.e42404. eCollection 2025 Feb 15.

ABSTRACT

This study presents a semi-automated approach for assessing water quality in the Sundarbans, a critical and vulnerable ecosystem, using machine learning (ML) models integrated with field and remotely-sensed data. Key water quality parameters-Sea Surface Temperature (SST), Total Suspended Solids (TSS), Turbidity, Salinity, and pH-were predicted through ML algorithms and interpolated using the Empirical Bayesian Kriging (EBK) model in ArcGIS Pro. The predictive framework leverages Google Earth Engine (GEE) and AutoML, utilizing deep learning libraries to create dynamic, adaptive models that enhance prediction accuracy. Comparative analyses showed that ML-based models effectively captured spatial and temporal variations, aligning closely with field measurements. This integration provides a more efficient alternative to traditional methods, which are resource-intensive and less practical for large-scale, remote areas. Our findings demonstrate that this semi-automated technique is a valuable tool for continuous water quality monitoring, particularly in ecologically sensitive areas with limited accessibility. The approach also offers significant applications for climate resilience and policy-making, as it enables timely identification of deteriorating water quality trends that may impact biodiversity and ecosystem health. However, the study acknowledges limitations, including the variability in data availability and the inherent uncertainties in ML predictions for dynamic water systems. Overall, this research contributes to the advancement of water quality monitoring techniques, supporting sustainable environmental management practices and the resilience of the Sundarbans against emerging climate challenges.

PMID:39981364 | PMC:PMC11840191 | DOI:10.1016/j.heliyon.2025.e42404

Front Artif Intell. 2025 Feb 6;8:1508664. doi: 10.3389/frai.2025.1508664. eCollection 2025.

ABSTRACT

In marine security and surveillance, accurately identifying ships and ship ports from satellite imagery remains a critical challenge due to the inefficiencies and inaccuracies of conventional approaches. The proposed method uses an enhanced YOLO (You Only Look Once) model, a robust real-time object detection method. The method involves training the YOLO model on an extensive collection of annotated satellite images to detect ships and ship ports accurately. The proposed system delivers a precision of 86% compared to existing methods; this approach is designed to allow for real-time deployment in the context of resource-constrained environments, especially with a Jetson Nano edge device. This deployment will ensure scalability, efficient processing, and reduced reliance on central computing resources, making it especially suitable for maritime settings in which real-time monitoring is vital. The findings of this study, therefore, point out the practical implications of this improved YOLO model for maritime surveillance: offering a scalable and efficient solution to strengthen maritime security.

PMID:39981193 | PMC:PMC11839658 | DOI:10.3389/frai.2025.1508664

Front Med (Lausanne). 2025 Feb 6;12:1506363. doi: 10.3389/fmed.2025.1506363. eCollection 2025.

ABSTRACT

BACKGROUND: Pulmonary embolism (PE) is a common and potentially fatal condition. Timely and accurate risk assessment in patients with acute deep vein thrombosis (DVT) is crucial. This study aims to develop a deep learning-based, precise, and efficient PE risk prediction model (PE-Mind) to overcome the limitations of current clinical tools and provide a more targeted risk evaluation solution.

METHODS: We analyzed clinical data from patients by first simplifying and organizing the collected features. From these, 37 key clinical features were selected based on their importance. These features were categorized and analyzed to identify potential relationships. Our prediction model uses a convolutional neural network (CNN), enhanced with three custom-designed modules for better performance. To validate its effectiveness, we compared this model with five commonly used prediction models.

RESULTS: PE-Mind demonstrated the highest accuracy and reliability, achieving 0.7826 accuracy and an area under the receiver operating characteristic curve of 0.8641 on the prospective test set, surpassing other models. Based on this, we have also developed a Web server, PulmoRiskAI, for real-time clinician operation.

CONCLUSION: The PE-Mind model improves prediction accuracy and reliability for assessing PE risk in acute DVT patients. Its convolutional architecture and residual modules substantially enhance predictive performance.

PMID:39981086 | PMC:PMC11839595 | DOI:10.3389/fmed.2025.1506363

Front Med (Lausanne). 2025 Feb 6;12:1477351. doi: 10.3389/fmed.2025.1477351. eCollection 2025.

ABSTRACT

PURPOSE: This study analyzed the research trends in machine learning (ML) pertaining to myocardial infarction (MI) from 2008 to 2024, aiming to identify emerging trends and hotspots in the field, providing insights into the future directions of research and development in ML for MI. Additionally, it compared the contributions of various countries, authors, and agencies to the field of ML research focused on MI.

METHOD: A total of 1,036 publications were collected from the Web of Science Core Collection database. CiteSpace 6.3.R1, Bibliometrix, and VOSviewer were utilized to analyze bibliometric characteristics, determining the number of publications, countries, institutions, authors, keywords, and cited authors, documents, and journals in popular scientific fields. CiteSpace was used for temporal trend analysis, Bibliometrix for quantitative country and institutional analysis, and VOSviewer for visualization of collaboration networks.

RESULTS: Since the emergence of research literature on medical imaging and machine learning (ML) in 2008, interest in this field has grown rapidly, particularly since the pivotal moment in 2016. The ML and MI domains, represented by China and the United States, have experienced swift development in research after 2015, albeit with the United States significantly outperforming China in research quality (as evidenced by the higher impact factors of journals and citation counts of publications from the United States). Institutional collaborations have formed, notably between Harvard Medical School in the United States and Capital Medical University in China, highlighting the need for enhanced cooperation among domestic and international institutions. In the realm of MI and ML research, cooperative teams led by figures such as Dey, Damini, and Berman, Daniel S. in the United States have emerged, indicating that Chinese scholars should strengthen their collaborations and focus on both qualitative and quantitative development. The overall direction of MI and ML research trends toward Medicine, Medical Sciences, Molecular Biology, and Genetics. In particular, publications in "Circulation" and "Computers in Biology and Medicine" from the United States hold prominent positions in this study.

CONCLUSION: This paper presents a comprehensive exploration of the research hotspots, trends, and future directions in the field of MI and ML over the past two decades. The analysis reveals that deep learning is an emerging research direction in MI, with neural networks playing a crucial role in early diagnosis, risk assessment, and rehabilitation therapy.

PMID:39981082 | PMC:PMC11839716 | DOI:10.3389/fmed.2025.1477351

MethodsX. 2025 Jan 25;14:103185. doi: 10.1016/j.mex.2025.103185. eCollection 2025 Jun.

ABSTRACT

Diabetic Retinopathy (DR), a diabetes-related eye condition, damages retinal blood vessels and can lead to vision loss if undetected early. Precise diagnosis is challenging due to subtle, varied symptoms. While classical deep learning (DL) models like CNNs and ResNet's are widely used, they face resource and accuracy limitations. Quantum computing, leveraging quantum mechanics, offers revolutionary potential for faster problem-solving across fields like cryptography, optimization, and medicine. This research introduces QuantumNet, a hybrid model combining classical DL and quantum transfer learning to enhance DR detection. QuantumNet demonstrates high accuracy and resource efficiency, providing a transformative solution for DR detection and broader medical imaging applications. The method is as follows:•Evaluate three classical deep learning models-CNN, ResNet50, and MobileNetV2-using the APTOS 2019 blindness detection dataset on Kaggle to identify the best-performing model for integration.•QuantumNet combines the best-performing classical DL model for feature extraction with a variational quantum classifier, leveraging quantum transfer learning for enhanced diagnostics, validated statistically and on Google Cirq using standard metrics.•QuantumNet achieves 94.11 % accuracy, surpassing classical DL models and prior research by 11.93 percentage points, demonstrating its potential for accurate, efficient DR detection and broader medical imaging applications.

PMID:39981059 | PMC:PMC11840206 | DOI:10.1016/j.mex.2025.103185

Health Sci Rep. 2025 Feb 19;8(2):e70200. doi: 10.1002/hsr2.70200. eCollection 2025 Feb.

ABSTRACT

BACKGROUND AND AIMS: Artificial intelligence (AI) has become integral to medical research, impacting various aspects such as data analysis, writing assistance, and publishing. This paper explores the multifaceted influence of AI on the process of writing medical research papers, encompassing data analysis, ethical considerations, writing assistance, and publishing efficiency.

METHODS: The review was conducted following the PRISMA guidelines; a comprehensive search was performed in Scopus, PubMed, EMBASE, and MEDLINE databases for research publications on artificial intelligence in medical research published up to October 2023.

RESULTS: AI facilitates the writing process by generating drafts, offering grammar and style suggestions, and enhancing manuscript quality through advanced models like ChatGPT. Ethical concerns regarding content ownership and potential biases in AI-generated content underscore the need for collaborative efforts among researchers, publishers, and AI creators to establish ethical standards. Moreover, AI significantly influences data analysis in healthcare, optimizing outcomes and patient care, particularly in fields such as obstetrics and gynecology and pharmaceutical research. The application of AI in publishing, ranging from peer review to manuscript quality control and journal matching, underscores its potential to streamline and enhance the entire research and publication process. Overall, while AI presents substantial benefits, ongoing research, and ethical guidelines are essential for its responsible integration into the evolving landscape of medical research and publishing.

CONCLUSION: The integration of AI in medical research has revolutionized efficiency and innovation, impacting data analysis, writing assistance, publishing, and others. While AI tools offer significant benefits, ethical considerations such as biases and content ownership must be addressed. Ongoing research and collaborative efforts are crucial to ensure responsible and transparent AI implementation in the dynamic landscape of medical research and publishing.

PMID:39980823 | PMC:PMC11839394 | DOI:10.1002/hsr2.70200

Front Plant Sci. 2025 Feb 3;15:1496801. doi: 10.3389/fpls.2024.1496801. eCollection 2024.

ABSTRACT

Accurate counting of crop plants is essential for agricultural science, particularly for yield forecasting, field management, and experimental studies. Traditional methods are labor-intensive and prone to errors. Unmanned Aerial Vehicle (UAV) technology offers a promising alternative; however, varying UAV altitudes can impact image quality, leading to blurred features and reduced accuracy in early maize seedling counts. To address these challenges, we developed RC-Dino, a deep learning methodology based on DINO, specifically designed to enhance the precision of seedling counts from UAV-acquired images. RC-Dino introduces two innovative components: a novel self-calibrating convolutional layer named RSCconv and an adaptive spatial feature fusion module called ASCFF. The RSCconv layer improves the representation of early maize seedlings compared to non-seedling elements within feature maps by calibrating spatial domain features. The ASCFF module enhances the discriminability of early maize seedlings by adaptively fusing feature maps extracted from different layers of the backbone network. Additionally, transfer learning was employed to integrate pre-trained weights with RSCconv, facilitating faster convergence and improved accuracy. The efficacy of our approach was validated using the Early Maize Seedlings Dataset (EMSD), comprising 1,233 annotated images of early maize seedlings, totaling 83,404 individual annotations. Testing on this dataset demonstrated that RC-Dino outperformed existing models, including DINO, Faster R-CNN, RetinaNet, YOLOX, and Deformable DETR. Specifically, RC-Dino achieved improvements of 16.29% in Average Precision (AP) and 8.19% in Recall compared to the DINO model. Our method also exhibited superior coefficient of determination (R²) values across different datasets for seedling counting. By integrating RSCconv and ASCFF into other detection frameworks such as Faster R-CNN, RetinaNet, and Deformable DETR, we observed enhanced detection and counting accuracy, further validating the effectiveness of our proposed method. These advancements make RC-Dino particularly suitable for accurate early maize seedling counting in the field. The source code for RSCconv and ASCFF is publicly available at https://github.com/collapser-AI/RC-Dino, promoting further research and practical applications.

PMID:39980762 | PMC:PMC11841422 | DOI:10.3389/fpls.2024.1496801

Genome Biol. 2025 Feb 20;26(1):34. doi: 10.1186/s13059-025-03503-y.

ABSTRACT

Spatially resolved transcriptomics (SRT) data provide critical insights into gene expression patterns within tissue contexts, necessitating effective methods for identifying spatial domains. We introduce stDyer, an end-to-end deep learning framework for spatial domain clustering in SRT data. stDyer combines Gaussian Mixture Variational AutoEncoder with graph attention networks to learn embeddings and perform clustering. Its dynamic graphs adaptively link units based on Gaussian Mixture assignments, improving clustering and producing smoother domain boundaries. stDyer's mini-batch strategy and multi-GPU support facilitate scalability to large datasets. Benchmarking against state-of-the-art tools, stDyer demonstrates superior performance in spatial domain clustering, multi-slice analysis, and large-scale dataset handling.

PMID:39980033 | DOI:10.1186/s13059-025-03503-y

BMC Med Educ. 2025 Feb 20;25(1):273. doi: 10.1186/s12909-024-06552-2.

ABSTRACT

BACKGROUND: The integration of artificial intelligence (AI) into medical education is poised to revolutionize teaching, learning, and clinical practice. However, successful implementation of AI-based tools in medical curricula faces several challenges, particularly in resource-limited settings like Pakistan, where technological and institutional barriers remain significant. This study aimed to evaluate knowledge, attitudes, and practices of medical students and faculty regarding AI in medical education, and explore the perceptions and key barriers regarding strategies for effective AI integration.

METHODS: A concurrent mixed-methods study was conducted over six months (July 2023 to January 2024) at a tertiary care medical college in Pakistan. The quantitative component utilized a cross-sectional design, with 236 participants (153 medical students and 83 faculty members) completing an online survey. Mean composite scores for knowledge, attitudes, and practices were analyzed using non-parametric tests. The qualitative component consisted of three focus group discussions with students and six in-depth interviews with faculty. Thematic analysis was performed to explore participants' perspectives on AI integration.

RESULTS: Majority of participants demonstrated a positive attitude towards AI integration. Faculty had significantly higher mean attitude scores compared to students (3.95 ± 0.63 vs. 3.81 ± 0.75, p = 0.040). However, no statistically significant differences in knowledge (faculty: 3.53 ± 0.66, students: 3.55 ± 0.73, p = 0.870) or practices (faculty: 3.19 ± 0.87, students: 3.23 ± 0.89, p = 0.891) were found. Older students reported greater self-perceived knowledge (p = 0.010) and more positive attitudes (p = 0.016) towards AI, while male students exhibited higher knowledge scores than females (p = 0.025). Qualitative findings revealed key themes, including AI's potential to enhance learning and research, concerns about over-reliance on AI, ethical issues surrounding privacy and confidentiality, and the need for institutional support. Faculty emphasized the importance of training to equip educators with the necessary skills to effectively integrate AI into their teaching.

CONCLUSIONS: This study highlights both the enthusiasm for AI integration and the significant barriers that must be addressed to successfully implement AI in medical education. Addressing technological constraints, providing faculty training, and developing ethical guidelines are critical steps toward fostering the responsible use of AI in medical curricula. These findings underscore the need for context-specific strategies, particularly in resource-limited settings, to ensure that medical students and educators are well-prepared for the future of healthcare.

PMID:39979912 | DOI:10.1186/s12909-024-06552-2

BMC Cancer. 2025 Feb 20;25(1):310. doi: 10.1186/s12885-025-13688-x.

ABSTRACT

BACKGROUND: Bladder cancer (BLCA) exists a profound molecular diversity, with basal and luminal subtypes having different prognostic and therapeutic outcomes. Traditional methods for molecular subtyping are often time-consuming and resource-intensive. This study aims to develop machine learning models using deep learning features from hematoxylin and eosin (H&E)-stained whole-slide images (WSIs) to predict basal and luminal subtypes in BLCA.

METHODS: RNA sequencing data and clinical outcomes were downloaded from seven public BLCA databases, including TCGA, GEO datasets, and the IMvigor210C cohort, to assess the prognostic value of BLCA molecular subtypes. WSIs from TCGA were used to construct and validate the machine learning models, while WSIs from Shanghai Tenth People's Hospital (STPH) and The Affiliated Guangdong Second Provincial General Hospital of Jinan University (GD2H) were used as external validations. Deep learning models were trained to obtained tumor patches within WSIs. WSI level deep learning features were extracted from tumor patches based on the RetCCL model. Support vector machine (SVM), random forest (RF), and logistic regression (LR) were developed using these features to classify basal and luminal subtypes.

RESULTS: Kaplan-Meier survival and prognostic meta-analyses showed that basal BLCA patients had significantly worse overall survival compared to luminal BLCA patients (hazard ratio = 1.47, 95% confidence interval: 1.25-1.73, P < 0.001). The LR model based on tumor patch features selected by Resnet50 model demonstrated superior performance, achieving an area under the curve (AUC) of 0.88 in the internal validation set, and 0.81 and 0.64 in the external validation sets from STPH and GD2H, respectively. This model outperformed both junior and senior pathologists in the differentiation of basal and luminal subtypes (AUC: 0.85, accuracy: 74%, sensitivity: 66%, specificity: 82%).

CONCLUSIONS: This study showed the efficacy of machine learning models in predicting the basal and luminal subtypes of BLCA based on the extraction of deep learning features from tumor patches in H&E-stained WSIs. The performance of the LR model suggests that the integration of AI tools into the diagnostic process could significantly enhance the accuracy of molecular subtyping, thereby potentially informing personalized treatment strategies for BLCA patients.

PMID:39979837 | DOI:10.1186/s12885-025-13688-x

J Imaging Inform Med. 2025 Feb 20. doi: 10.1007/s10278-025-01448-z. Online ahead of print.

ABSTRACT

Medical image segmentation is challenging due to the need for expert annotations and the variability of these manually created labels. Previous methods tackling label variability focus on 2D segmentation and single modalities, but reliable 3D multi-modal approaches are necessary for clinical applications such as in oncology. In this paper, we propose a framework for generating reliable and unbiased labels with minimal radiologist input for supervised 3D segmentation, reducing radiologists' efforts and variability in manual labeling. Our framework generates AI-assisted labels through a two-step process involving 3D multi-modal unsupervised segmentation based on feature clustering and semi-supervised refinement. These labels are then compared against traditional expert-generated labels in a downstream task consisting of 3D multi-modal bone tumor segmentation. Two 3D-Unet models are trained, one with manually created expert labels and the other with AI-assisted labels. Following this, a blind evaluation is performed on the segmentations of these two models to assess the reliability of training labels. The framework effectively generated accurate segmentation labels with minimal expert input, achieving state-of-the-art performance. The model trained with AI-assisted labels outperformed the baseline model in 61.67% of blind evaluations, indicating the enhancement of segmentation quality and demonstrating the potential of AI-assisted labeling to reduce radiologists' workload and improve label reliability for 3D multi-modal bone tumor segmentation. The code is available at https://github.com/acurtovilalta/3D_LabelGeneration .

PMID:39979760 | DOI:10.1007/s10278-025-01448-z

J Imaging Inform Med. 2025 Feb 20. doi: 10.1007/s10278-024-01380-8. Online ahead of print.

ABSTRACT

Clinical parameters measured from gated single-photon emission computed tomography myocardial perfusion imaging (SPECT MPI) have value in predicting cardiac resynchronization therapy (CRT) patient outcomes, but still show limitations. The purpose of this study is to combine clinical variables, features from electrocardiogram (ECG), and parameters from assessment of cardiac function with polar maps from gated SPECT MPI through deep learning (DL) to predict CRT response. A total of 218 patients who underwent rest-gated SPECT MPI were enrolled in this study. CRT response was defined as an increase in left ventricular ejection fraction (LVEF) > 5% at a 6-month follow-up. A DL model was constructed by combining a pre-trained VGG16 model and a multilayer perceptron. Two modalities of data were input to the model: polar map images from SPECT MPI and tabular data from clinical features, ECG parameters, and SPECT-MPI-derived parameters. Gradient-weighted class activation mapping (Grad-CAM) was applied to the VGG16 model to provide explainability for the polar maps. For comparison, four machine learning (ML) models were trained using only the tabular features. Modeling was performed on 218 patients who underwent CRT implantation with a response rate of 55.5% (n = 121). The DL model demonstrated average AUC (0.83), accuracy (0.73), sensitivity (0.76), and specificity (0.69) surpassing ML models and guideline criteria. Guideline recommendations achieved accuracy (0.53), sensitivity (0.75), and specificity (0.26). The DL model trended towards improvement over the ML models, showcasing the additional predictive benefit of utilizing SPECT MPI polar maps. Incorporating additional patient data directly in the form of medical imagery can improve CRT response prediction.

PMID:39979759 | DOI:10.1007/s10278-024-01380-8

Mol Imaging Biol. 2025 Feb 20. doi: 10.1007/s11307-025-01990-w. Online ahead of print.

ABSTRACT

PURPOSE: The goal of this study is to create a novel framework for identifying MSI status in colorectal cancer using advanced radiomics and deep learning strategies, aiming to enhance clinical decision-making and improve patient outcomes in oncology.

PROCEDURES: The study utilizes histopathological slide images from the NCT-CRC-HE-100 K and PAIP 2020 databases. Key procedures include self-attentive adversarial stain normalization for data standardization, tumor delineation via a Slimmable Transformer, and radiomics feature extraction using a hybrid quantum-classical neural network.

RESULTS: The proposed system reaches 99% accuracy when identifying colorectal cancer MSI status. It shows the model is good at telling the difference between MSI and MSS tumors and can be used in real medical care for cancer.

CONCLUSIONS: Our research shows that the new system improves colorectal cancer MSI status determination better than previous methods. Our optimized processing technology works better than other methods to divide and analyze tissue features making the system good for improving patient care decisions.

PMID:39979579 | DOI:10.1007/s11307-025-01990-w

Endocrine. 2025 Feb 20. doi: 10.1007/s12020-025-04196-w. Online ahead of print.

ABSTRACT

PURPOSE: This study explores a self-learning method as an auxiliary approach in residency training for distinguishing between benign and malignant thyroid nodules.

METHODS: Conducted from March to December 2022, internal medicine residents underwent three repeated learning sessions with a "learning set" comprising 3000 thyroid nodule images. Diagnostic performances for internal medicine residents were assessed before the study, after every learning session, and for radiology residents before and after one-on-one education, using a "test set," comprising 120 thyroid nodule images. Finally, all residents repeated the same test using artificial intelligence computer-assisted diagnosis (AI-CAD).

RESULTS: Twenty-one internal medicine and eight radiology residents participated. Initially, internal medicine residents had a lower area under the receiver operating characteristic curve (AUROC) than radiology residents (0.578 vs. 0.701, P < 0.001), improving post-learning (0.578 to 0.709, P < 0.001) to a comparable level with radiology residents (0.709 vs. 0.735, P = 0.17). Further improvement occurred with AI-CAD for both group (0.709 to 0.755, P < 0.001; 0.735 to 0.768, P = 0.03).

CONCLUSION: The proposed iterative self-learning method using a large volume of ultrasonographic images can assist beginners, such as residents, in thyroid imaging to differentiate benign and malignant thyroid nodules. Additionally, AI-CAD can improve the diagnostic performance across varied levels of experience in thyroid imaging.

PMID:39979566 | DOI:10.1007/s12020-025-04196-w