Chest X-rays (CXRs) are among the most frequently performed imaging examinations worldwide and remain central to evaluating thoracic disease. Their interpretation, however, can be challenging in high-volume clinical environments, as it requires expertise and careful assessment of both the presence and severity of abnormalities. Although deep learning has enabled significant progress in automated disease detection and report generation, these tasks are often developed in isolation, preventing models from leveraging complementary textual and visual cues. Moreover, existing triage approaches typically rely on coarse categorizations rather than continuous, interpretable severity estimation. To address these limitations, we propose the Multimodal Triage Network (MTN), a unified model that jointly predicts disease captions and continuous severity scores by integrating image and text modalities. MTN employs a dual-stream encoder using ResNet-18 for visual features and BioBERT for textual features, followed by crossmodal attention to align and fuse information across domains. A multi-task decoder then generates structured natural-language captions describing the findings while simultaneously predicting a continuous severity score. Experiments on a curated subset of the MIMIC-CXR-JPG dataset show that MTN substantially outperforms its unimodal counterparts, achieving an F1 score of 0.9746 for captioning, a severity prediction MSE of 0.0448, and a classification accuracy of 0.9861. By ranking cases based on predicted severity, the model provides an effective mechanism for automated triage and workflow prioritization. Overall, MTN offers a clinically meaningful framework that combines interpretability, severity awareness, and high predictive performance in a single multimodal system.

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, and Computed Tomography (CT) is a primary imaging tool for screening and followup assessment. After pulmonary nodule detection, radiologists manually assess anatomical location, diameter, margin characteristics, and attenuation type to support risk assessment and clinical decision-making. However, this post-detection workflow is time-consuming and can be affected by inter-observer variability. Existing Artificial Intelligence methods often focus on isolated tasks, limiting their use as a unified, clinically grounded interpretation framework. This study presents FZ-VLM, a two-stage Florence-Zephyr Vision Language Model framework for unified structured pulmonary nodule characterization in lung CT. The framework uses a fine-tuned Florence-2 model to extract radiological attributes from expert-annotated 2D axial CT slices, while a Zephyr-7B model uses these attributes to generate nodule descriptions, follow-up recommendations, and longitudinal analyses. Results showed that the Stage 1 model achieved 77.18% accuracy for anatomical location, 67.96% accuracy for margin characteristics, and 79.13% accuracy for attenuation type, with a Mean Absolute Error of 2.58 mm for diameter estimation, outperforming evaluated GPT-4-based baselines as well as the human baseline. Expert radiologist evaluation of Stage 2 showed 93.9% accuracy, 98.6% completeness score, 76.1% clinical relevance, and an overall score of 89.5%. Safety analysis showed that most outputs were clinically safe, although some follow-up recommendations still required expert review. To the best of our knowledge, this study presents the first two-stage Vision-Language Model framework for structured nodule characterization and clinical decision-making.

BACKGROUND:Transperineal ultrasound (TPUS) is a valuable imaging tool for evaluating patients with a variety of pelvic floor disorders, including pelvic organ prolapse (POP). Currently, calculating measurements of anatomical structures and relationships as well as extracting the mid-sagittal (MS) plane of 2D and 3D ultrasound images are obtained manually, which is a time-consuming process and requires a reviewer with prior training in pelvic floor US interpretation. The need for manual analysis of ultrasound images has limited the broader adoption of TPUS for evaluating pelvic floor disorders in both research and clinical practice. An automated segmentation and plane extraction method would improve the ability to easily quantify pelvic anatomy relevant to pelvic floor disorders and improve the efficiency and reproducibility of POP diagnosis and treatment. PURPOSE:To develop a fast, reproducible, and automated method of acquiring the MS plane, plane of minimal hiatal dimensions (PMHD), and segmentations of the pelvic floor organs from 3D TPUS images. METHODS:Our method used a nnU-Net segmentation model to segment structures of interest in the 3D TPUS images. The model segmented the pubis symphysis (PS), urethra, bladder, rectum, rectal ampulla, and anorectal angle (ANA). The segmented output was then fed into a heuristics-based method to determine the PS and ANA to extract the MS plane and PMHD automatically. We used a dataset consisting of 161 3D TPUS images from 104 patients. 89 of the volumes were acquired in a resting state and 72 during the Valsalva maneuver. The segmentation and plane extraction algorithms were evaluated by comparing the results with manual segmentations and manual plane extraction methods using the dice similarity coefficients (DSC), mean absolute surface distance (MAD), and absolute angle difference (AAD), respectively. The Wilcoxon-signed rank statistical test was used with Bonferroni-correction to p < 0.01. Cohen effect size was used for comparing model results. RESULTS:The nnU-Net segmentation model reported an average DSC(%) of 70.4%, 58.5%, 57.1%, 48.9%, 39.0%, and 19.8% for bladder, rectum, PS, urethra, ANA, and rectal ampulla respectively. The nnU-Net segmentation model achieved significantly higher DSC (p < 0.01) for the urethra and rectum than all other tested models. Across all metrics, the nnU-Net segmentation model achieved an average effect size of 0.3, 0.5, 0.7, and 0.8 compared to a 3D ResNet34 + U-Net, 3D U-Net, 2D U-Net, and Attention 3D U-Net model, respectively. The average AADs between the automatically calculated plane slices and manually estimated planes dataset for the MS plane and PMHD were 3.8° and 2.4°, respectively. The PS and ANA segmentation centroids were used to calculate the MS plane and PMHD and they had distance errors of 3.6 mm and 4.4 mm. CONCLUSIONS:We developed an automated 3D segmentation and multiple plane extraction method of female pelvic floor 3D US images. Our method extracts the MS plane and PMHD from 3D US images. The proposed algorithm pipeline can improve the efficiency and reproducibility of TPUS analysis for pelvic floor disorder diagnosis and treatment.

Lung cancer remains the leading cause of cancer-related mortality worldwide, underscoring the need for efficient and reliable diagnostic workflows. Although computed tomography (CT) is widely used for early detection, the post-detection interpretation of tumor-bearing slices is time-consuming and prone to interobserver variability. In this study, we investigate the use of vision-language models (VLMs) as a post-detection tool for structured clinical interpretation of lung CT images. We fine-tuned Florence-2 (Large) using paired image-question-answer triplets curated from the National Lung Screening Trial (NLST) dataset, targeting four clinically relevant attributes of the tumour: anatomical location, longest diameter, margin characteristics, and attenuation type. For comparative analysis, the model was evaluated against both fine-tuned and pre-trained GPT-4 baselines using the same test protocol. The fine-tuned Florence-2 model achieved strong performance across all tasks, including 74.3% accuracy for anatomical location, a mean absolute error of 2.58 mm for diameter estimation, 67.96% accuracy for margin classification, and 78.12% accuracy for attenuation classification. This comparative analysis demonstrated that Florence-2 consistently outperformed both fine-tuned and pre-trained GPT-4 models across all evaluated tasks. These results indicate that domain-specific fine-tuning and architectural alignment play a critical role in structured, spatially precise interpretation of medical images. This work demonstrates the effectiveness of adapting vision-language models for clinically structured post-detection interpretation of lung CT scans, supporting accurate and consistent radiological assessment.

Purpose: Increased number of chest radiograph (CXR) scans create a triage bottleneck, queueing urgent examinations behind routine ones. Existing AI tools are predominantly unimodal binary classifiers lacking severity awareness, and multimodal systems are rarely benchmarked against expert radiologists. To this end, we developed a multimodal deep learning framework for joint severity triage, pathology detection, and native visual explanation. Approach: We propose the cross-modal triage network (CMTN), fusing a Swin Transformer V2 visual encoder with a PubMedBERT text encoder via gated cross-attention. The CMTN was trained on 34,639 image-text pairs (12,489 patients) from MIMIC-CXR-JPG, optimizing an ordinal focal loss for four-tier severity triage and binary cross-entropy for 14 pathologies. Beyond quantitative benchmarking, attention heatmaps were evaluated against a blinded expert radiologist in a two-phase clinical audit comparing model triage output to expert severity assessment (100 cases) and grading spatial-semantic concordance (116 heatmaps). Results: The CMTN achieved strong ordinal agreement with reference labels (quadratic weighted kappa [QWK] = 0.9341, 95% CI: 0.9219 to 0.9449) and macro-AUROC of 0.9970 across 14 pathologies, with 34 ms latency, outperforming the state-of-the-art BioViL multimodal baseline (QWK = 0.7679). However, the blinded Phase I clinical audit revealed substantially lower agreement with genuine radiologist judgment (QWK = 0.1399). Phase II found 54.3% of heatmaps achieved clinically acceptable spatial localization. Conclusions: The CMTN demonstrated an efficient multimodal architecture for CXR triage. The divergence between algorithmic and radiologist agreement demonstrates that benchmark performance against NLP-derived labels is insufficient, highlighting the need for radiologist-labeled ground truth before clinical deployment.

Large volumes of medical data remain underutilized because centralizing distributed data is often infeasible due to strict privacy regulations and institutional constraints. In addition, models trained in centralized settings frequently fail to generalize across clinical sites because of heterogeneity in imaging protocols and continuously evolving data distributions arising from differences in scanners, acquisition parameters, and patient populations. Federated learning offers a promising solution by enabling collaborative model training without sharing raw data. However, incorporating differential privacy into federated learning, while essential for privacy guarantees, often leads to degraded accuracy, unstable convergence, and reduced generalization. In this work, we propose an adaptive differentially private federated learning (ADP-FL) framework for medical image segmentation that dynamically adjusts privacy mechanisms to better balance the privacy-utility trade-off. The proposed approach stabilizes training, significantly improves Dice scores and segmentation boundary quality, and maintains rigorous privacy guarantees. We evaluated ADP-FL across diverse imaging modalities and segmentation tasks, including skin lesion segmentation in dermoscopic images, kidney tumor segmentation in 3D CT scans, and brain tumor segmentation in multi-parametric MRI. Compared with conventional federated learning and standard differentially private federated learning, ADP-FL consistently achieves higher accuracy, improved boundary delineation, faster convergence, and greater training stability, with performance approaching that of non-private federated learning under the same privacy budgets. These results demonstrate the practical viability of ADP-FL for high-performance, privacy-preserving medical image segmentation in real-world federated settings.

Metabolic dysfunction-associated steatotic liver disease (MASLD) is the most prevalent chronic liver disease in the United States and Canada, affecting at least 25% of adults. Early diagnosis is critical, as untreated MASLD can progress to metabolic dysfunction-associated steatohepatitis (MASH), liver fibrosis and ultimately to cirrhosis. Despite this burden, accessible and non-invasive screening tools for liver disease remain limited. Point-of-care ultrasound (POCUS) offers a cost-effective and widely available imaging modality in primary care settings; however, its clinical utility is constrained by low spatial resolution, variable image quality, and strong dependence on operator expertise for acquiring diagnosticquality images. To address these limitations, we are developing a suite of AI-based tools to provide image guidance and decision support for POCUS users. In this work, we introduce an automated segmentation framework capable of identifying the liver in both high- and low-quality abdominal POCUS images, with the potential to assist in liver disease assessment. Specifically, we propose a U-Net-based model to segment the liver and acoustic shadow regions in POCUS images. In parallel, we implement a ResNet-based convolutional neural network (CNN) to classify image quality as either suitable or unsuitable for clinical diagnosis. The segmentation model was trained and evaluated on a dataset of 8,285 POCUS images from 149 patients, while an independent dataset of 1,116 POCUS images from 13 patients was used for diagnostic image quality classification. The U-Net achieved average Dice similarity coefficients (DSC) of 68.1% and 52.6% for liver and acoustic shadow segmentation, respectively. The CNN image quality classifier achieved an accuracy of 70.0%, with precision, recall, specificity, and F1-score of 59.8%, 82.9%, 61.0%, and 69.5%, respectively.

BACKGROUND:Federated learning (FL) facilitates collaborative model training across multiple institutions while preserving privacy by avoiding the sharing of raw data, a critical consideration in medical imaging applications. Despite its potential, FL faces challenges such as high-dimensional data, heterogeneity among datasets from different centers, and resource constraints, which limit its efficiency and effectiveness in healthcare settings. PURPOSE:This study aims to present a novel adaptive FL framework to address the challenges of data heterogeneity and resource constraints in medical imaging. The proposed framework is designed to optimize computational efficiency, enhance training processes, improve model performance, and ensure robustness against non-independent and identically distributed (non-IID) data across decentralized data sources. METHODS:The proposed adaptive FL framework addresses the challenges of high-dimensional data and heterogeneity in nonuniform and decentralized data sources through a key innovation. First, Federated incremental principal component analysis (FIPCA) achieves privacy-preserving dimensionality reduction by aggregating local scatter matrices and means from participating centers, enabling the computation of a global PCA model. This process ensures data alignment across centers, mitigates heterogeneity, and significantly reduces computational complexity. We evaluated the framework's ability to generalize across institutions in a cross-site classification task distinguishing clinically significant prostate cancer (csPCa) from non-csPCa. This assessment used 1500 T2-weighted (T2W) prostate MRI images from three institutions, where two centers (800 + 350 cases) were used for training and validation, and one center (350 cases) served as an independent test site. RESULTS:The proposed method significantly reduced the number of global training rounds from 200 to 38, achieving a 98% reduction in energy consumption compared to the standard FedAvg algorithm. The effective use of FIPCA for dimensionality reduction enhanced generalizability, while adaptive early stopping prevented overfitting, leading to an improvement in model performance, with the area under the curve (AUC) on the unseen test center increasing from 0.68 to 0.73 (95 % CI 0.70 - 0.77) on the test center's data. Additionally, the method demonstrated improved sensitivity and specificity, indicating superior classification performance. The integration of FIPCA accelerated convergence by reducing data dimensionality, while the adaptive early-stopping mechanism further optimized resource utilization and prevented overfitting. CONCLUSIONS:Our adaptive FL approach efficiently handles large, heterogeneous medical imaging data, reducing training time and computational overhead, while improving model accuracy. The substantial reduction in energy consumption and accelerated convergence make it suitable for real-world healthcare settings.

Objective:The purpose of this research was to examine the feasibility of utilizing generative adversarial networks (GANs) to generate accurate pseudo-CT images for dogs. Methods:This study used head standard CT images and T1-weighted transverse with contrast 3-D fast spoiled gradient echo head MRI images from 45 nonbrachycephalic dogs that received treatment between 2014 and 2023. Two conditional GANs (CGANs), one with a U-Net generator and a PatchGAN discriminator and another with a residual neural network (ResNet) U-Net generator and ResNet discriminator were used to generate the pseudo-CT images. Results:The CGAN with a ResNet U-Net generator and ResNet discriminator had an average mean absolute error of 109.5 ± 153.7 HU, average peak signal-to-noise ratio of 21.2 ± 4.31 dB, normalized mutual information of 0.89 ± 0.05, and dice similarity coefficient of 0.91 ± 0.12. The dice similarity coefficient for the bone was 0.71 ± 0.17. Qualitative results indicated that the most common ranking was "slightly similar" for both models. The CGAN with a ResNet U-Net generator and ResNet discriminator produced more accurate pseudo-CT images than the CGAN with a U-Net generator and PatchGAN discriminator. Conclusions:The study concludes that CGAN can generate relatively accurate pseudo-CT images but suggests exploring alternative GAN extensions. Clinical Relevance:Implementing generative learning into veterinary radiation therapy planning demonstrates the potential to reduce imaging costs and time.

Compared to conventional two-dimensional (2D) ultrasound, three-dimensional (3D) ultrasound (US) images are a more sensitive alternative for monitoring the size and shape of neonatal cerebral lateral ventricles for monitoring intraventricular hemorrhaging (IVH). The ventricles must be segmented by an expert to estimate the ventricular volume, which can be time-consuming and difficult to obtain. In this paper, we describe a scribble-based weakly supervised segmentation method that trains only on non-expert drawn scribbles. We trained and tested two segmentation methods, a vanilla 3D U-Net benchmark and a weakly supervised learning for medical image segmentation (WSL4MIS) method built into the 3D U-Net model, using 56 3D US images. We performed two experiments, the first where models were trained, validated, and tested on 25/5/15 images respectively, and a second where models were trained, validated, and tested on 36/5/15 images respectively. For the first experiment, the 3D U-Net and WSL4MIS achieved a mean +/- standard deviation and Dice similarity coefficients (DSC) of 40.4 +/- 9.4% and 42.8 +/- 7.9%, respectively. The second experiment yielded a DSC of 40.4 +/- 7.2% and 44.8 +/- 9.5% for 3D U-Net and WSL4MIS, respectively. For both experiments, the WSL4MIS method had a higher mean DSC and lower standard deviation than the baseline 3D U-Net when both used scribbles for training. When trained on 36 images and scribbles, the WSL4MIS achieved statistically significant DSC values compared to the 3D U-Net model based on the Wilcoxon signed-rank test.

Purpose:Semantic segmentation in high-resolution, histopathology whole slide images (WSIs) is an important fundamental task in various pathology applications. Convolutional neural networks (CNN) are the state-of-the-art approach for image segmentation. A patch-based CNN approach is often employed because of the large size of WSIs; however, segmentation performance is sensitive to the field-of-view and resolution of the input patches, and balancing the trade-offs is challenging when there are drastic size variations in the segmented structures. We propose a multiresolution semantic segmentation approach, which is capable of addressing the threefold trade-off between field-of-view, computational efficiency, and spatial resolution in histopathology WSIs.Approach:We propose a two-stage multiresolution approach for semantic segmentation of histopathology WSIs of mouse lung tissue and human placenta. In the first stage, we use four different CNNs to extract the contextual information from input patches at four different resolutions. In the second stage, we use another CNN to aggregate the extracted information in the first stage and generate the final segmentation masks.Results:The proposed method reported 95.6%, 92.5%, and 97.1% in our single-class placenta dataset and 97.1%, 87.3%, and 83.3% in our multiclass lung dataset for pixel-wise accuracy, mean Dice similarity coefficient, and mean positive predictive value, respectively.Conclusions:The proposed multiresolution approach demonstrated high accuracy and consistency in the semantic segmentation of biological structures of different sizes in our single-class placenta and multiclass lung histopathology WSI datasets. Our study can potentially be used in automated analysis of biological structures, facilitating the clinical research in histopathology applications.

Purpose Lung cancer is the second most common cancer and the leading cause of cancer death globally. Low dose computed tomography (LDCT) is the recommended imaging screening tool for the early detection of lung cancer. A fully automated computer-aided detection method for LDCT will greatly improve the existing clinical workflow. Most of the existing methods for lung detection are designed for high-dose CTs (HDCTs), and those methods cannot be directly applied to LDCTs due to domain shifts and inferior quality of LDCT images. In this work, we describe a semi-automated transfer learning-based approach for the early detection of lung nodules using LDCTs. Approach In this work, we developed an algorithm based on the object detection model, you only look once (YOLO) to detect lung nodules. The YOLO model was first trained on CTs, and the pre-trained weights were used as initial weights during the retraining of the model on LDCTs using a medical-to-medical transfer learning approach. The dataset for this study was from a screening trial consisting of LDCTs acquired from 50 biopsy-confirmed lung cancer patients obtained over 3 consecutive years (T1, T2, and T3). About 60 lung cancer patients' HDCTs were obtained from a public dataset. The developed model was evaluated using a hold-out test set comprising 15 patient cases (93 slices with cancerous nodules) using precision, specificity, recall, and F1-score. The evaluation metrics were reported patient-wise on a per-year basis and averaged for 3 years. For comparative analysis, the proposed detection model was trained using pre-trained weights from the COCO dataset as the initial weights. A paired t-test and chi-squared test with an alpha value of 0.05 were used for statistical significance testing. Results The results were reported by comparing the proposed model developed using HDCT pre-trained weights with COCO pre-trained weights. The former approach versus the latter approach obtained a precision of 0.982 versus 0.93 in detecting cancerous nodules, specificity of 0.923 versus 0.849 in identifying slices with no cancerous nodules, recall of 0.87 versus 0.886, and F1-score of 0.924 versus 0.903. As the nodule progressed, the former approach achieved a precision of 1, specificity of 0.92, and sensitivity of 0.930. The statistical analysis performed in the comparative study resulted in a p-value of 0.0054 for precision and a p-value of 0.00034 for specificity. Conclusions In this study, a semi-automated method was developed to detect lung nodules in LDCTs using HDCT pre-trained weights as the initial weights and retraining the model. Further, the results were compared by replacing HDCT pre-trained weights in the above approach with COCO pre-trained weights. The proposed method may identify early lung nodules during the screening program, reduce overdiagnosis and follow-ups due to misdiagnosis in LDCTs, start treatment options in the affected patients, and lower the mortality rate. (c) 2024 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI:10.1117/1.JMI.11.4.044502]

Due to superior soft tissue contrast afforded by magnetic resonance imaging (MRI), there is great potential for multi-parametric MRI (mpMRI) for the detection and eventual classification of renal masses (RMs). In this study, we investigated fully automated deep learning methods for RMs detection using T2-Weighted (T2W) spin-echo and two contrast-enhanced T1-Weighted gradient-echo-corticomedullary (T1W-CM), nephrographic-phase (T1W-NG), T1-Weighted In-phase (T1W-IP) and opposed-phase (T1W-OP) images. The dataset contained mpMRI images of 108 kidney cancer patients with an average size of renal mass of 24 ± 7.8 cm. In the first stage, kidneys were segmented using a 2D attention U-Net model, which was reported in a previous study. In the second stage, we tested five different state-of-the-art methods for RMs detections on mpMRI sequences. The model predictions were compared to manual annotations using precision, recall, specificity, and Dice Similarity Coefficient (DSC). The best-performing deep learning models were U-Net, U-Net++, and attention U-Net on the T2W, T1W-CM, and T1W-NG sequences respectively. Of the 5 mpMRI sequences, we also demonstrated that the T1W-CM is the most suitable for RMs detection. This automated detection of RMs in mpMRI sequences may be useful for the subsequent characterization of RMs in a fully automated artificial intelligence-based pipeline.

Placental lesions indicative of maternal vascular malperfusion (MVM) are associated with future cardiovascular disease (CVD) in women with placenta-mediated diseases of pregnancy. Early diagnosis of CVD can reduce morbidity, mortality, and healthcare costs. MVM lesions can be detected in placental histopathology images, providing a means for CVD risk screening postnatally. Deep learning approaches, such as convolutional neural networks (CNNs), have demonstrated high potential for automating histopathology image analysis. Given the large size of histopathology whole slide images (WSIs), a patch-based approach is often employed; however, labeling is typically only available at the WSI level. MVM lesions can present heterogeneously across the image, so assigning WSI labels to patches results in patch mislabeling. In this study, we propose a weakly supervised learning method for MVM lesion classification. Features were computed from the patches extracted from WSIs using the CNN-based Resnet18 pre-trained model. An attention-based network, using weakly supervised learning, has been developed to classify WSIs as MVM+/-. The model performance was assessed against a baseline model, which was a patch-based fully supervised model. The weakly supervised learning method had an accuracy of 87.4%, which was superior to the baseline model accuracy of 81.1%. This is a promising result that suggests that weakly supervised learning can help overcome patch labeling errors that arise from the heterogeneous presentation of histopathology features, such as MVM lesions, and only having WSI-level labels.

Magnetic resonance imaging (MRI) is well suited for Solid renal masses (SRMs) characterization (e.g., benign vs. malignant) due to its superior soft tissue contrast. Though renal mass detection and characterization using deep-learning (DL) methods have been extensively studied for computed tomography (CT) images, those same tasks are yet to be investigated on MRI images. SRMs need active surveillance as they comprise biologically diverse heterogeneous groups of benign or malignant masses. Among them, malignant clear cell renal carcinoma (ccRCC) is frequently aggressive. There are inter-observer and intra-observer differences in the assessment of SRMs by expert clinicians because of their experience and expertise. Therefore, it is essential to develop a machine learning-based noninvasive imaging diagnosis to distinguish SRMs as benign and malignant. Our retrospective study consisted of malignant (renal cell carcinoma- clear cell, papillary, and chromophobe) and benign (fat-poor angiomyolipoma-fpAML, oncocytomas) SRMs. We extracted first and second-order radiomics features from SRMs on T2W and T1W-CM MRI to train different machine learning (ML) models using the 5-fold cross-validation for benign vs malignant classification. The support vector machine (SVM) algorithm generated benign vs malignant classification accuracy of 90.00% with ROC-AUC of 76.19% on T2W MRI and the custom-designed multilayer perceptron model (MLP) model produced accuracy of 80.00% with ROC-AUC of 75.47% on T1W-CM MRI. Thus, ML-based radiomics features classification of SRMs extracted on MRI may be an alternative to biopsy using a non-invasive assessment of SRMs.

Background. ThePlease provide an email address for the corresponding author. risk of biochemical recurrence (BCR) after radiotherapy for localized prostate cancer (PCa) varies widely within standard risk groups. There's a need for low-cost tools to more robustly predict recurrence and personalize therapy. Radiomic features from pretreatment MRI show potential as noninvasive biomarkers for BCR prediction. Previous research has not fully combined radiomics with clinical and pathological data in predicting BCR of PCa patients after radiotherapy. Purpose. This study aims to predict 5-year BCR using radiomics from pretreatment T2W MRI and clinical-pathological data in PCa patients treated with radiation therapy, and to develop a unified model compatible with 1.5T and 3T MRI scanners. Methods. 150 T2W scans and clinical parameters were preprocessed. 120 cases were used for training and validation, and 30 for testing. Four distinct machine learning models were developed: Model 1 used radiomics, Model 2 used clinical and pathological data, Model 3 combined these via late fusion. Model 4 integrated radiomic and clinical-pathological data via early fusion . Results. Model 1 achieved an AUC of 0.73, while Model 2 had an AUC of 0.64 for predicting outcomes in 30 new test cases. Model 3, using late fusion, had an AUC of 0.69. Early fusion models showed promise: Model 4 reached an AUC of 0.84 highlighting the effectiveness of early fusion model. Conclusions. This study is the first to use fusion technique for predicting BCR in PCa patients following radiotherapy, using pre-treatment T2W MRI images and clinical-pathological data. Our methodology improves predictive accuracy by fusing radiomics with clinical-pathological information, even with a small dataset, and introduces the first unified model for both 1.5T and 3T MRI images.

Thoracic radiographs are an essential diagnostic tool in companion animal medicine and are frequently used as a part of routine workups in patients presenting for coughing, respiratory distress, cardiovascular diseases, and for staging of neoplasia. Quality control is a critical aspect of radiology practice in preventing misdiagnosis and ensuring consistent, accurate, and reliable diagnostic imaging. Implementing an effective quality control procedure in radiology can impact patient outcomes, facilitate clinical decision-making, and decrease healthcare costs. In this study, a machine learning-based quality classification model is suggested for canine and feline thoracic radiographs captured in both ventrodorsal and dorsoventral positions. The problem of quality classification was divided into collimation, positioning, and exposure, and then an automatic classification method was proposed for each based on deep learning and machine learning. We utilized a dataset of 899 radiographs of dogs and cats. Evaluations using fivefold cross-validation resulted in an F1 score and AUC score of 91.33 (95% CI: 88.37-94.29) and 91.10 (95% CI: 88.16-94.03), respectively. Results indicated that the proposed automatic quality classification has the potential to be implemented in radiology clinics to improve radiograph quality and reduce nondiagnostic images.

BACKGROUND:Three-dimensional (3D) ultrasound (US) imaging has shown promise in non-invasive monitoring of changes in the lateral brain ventricles of neonates suffering from intraventricular hemorrhaging. Due to the poorly defined anatomical boundaries and low signal-to-noise ratio, fully supervised methods for segmentation of the lateral ventricles in 3D US images require a large dataset of annotated images by trained physicians, which is tedious, time-consuming, and expensive. Training fully supervised segmentation methods on a small dataset may lead to overfitting and hence reduce its generalizability. Semi-supervised learning (SSL) methods for 3D US segmentation may be able to address these challenges but most existing SSL methods have been developed for magnetic resonance or computed tomography (CT) images. PURPOSE:To develop a fast, lightweight, and accurate SSL method, specifically for 3D US images, that will use unlabeled data towards improving segmentation performance. METHODS:We propose an SSL framework that leverages the shape-encoding ability of an autoencoder network to enforce complex shape and size constraints on a 3D U-Net segmentation model. The autoencoder created pseudo-labels, based on the 3D U-Net predicted segmentations, that enforces shape constraints. An adversarial discriminator network then determined whether images came from the labeled or unlabeled data distributions. We used 887 3D US images, of which 87 had manually annotated labels and 800 images were unlabeled. Training/validation/testing sets of 25/12/50, 25/12/25 and 50/12/25 images were used for model experimentation. The Dice similarity coefficient (DSC), mean absolute surface distance (MAD), and absolute volumetric difference (VD) were used as metrics for comparing to other benchmarks. The baseline benchmark was the fully supervised vanilla 3D U-Net while dual task consistency, shape-aware semi-supervised network, correlation-aware mutual learning, and 3D U-Net Ensemble models were used as state-of-the-art benchmarks with DSC, MAD, and VD as comparison metrics. The Wilcoxon signed-rank test was used to test statistical significance between algorithms for DSC and VD with the threshold being p < 0.05 and corrected to p < 0.01 using the Bonferroni correction. The random-access memory (RAM) trace and number of trainable parameters were used to compare the computing efficiency between models. RESULTS:Relative to the baseline 3D U-Net model, our shape-encoding SSL method reported a mean DSC improvement of 6.5%, 7.7%, and 4.1% with a 95% confidence interval of 4.2%, 5.7%, and 2.1% using image data splits of 25/12/50, 25/12/25, and 50/12/25, respectively. Our method only used a 1GB increase in RAM compared to the baseline 3D U-Net and required less than half the RAM and trainable parameters compared to the 3D U-Net ensemble method. CONCLUSIONS:Based on our extensive literature survey, this is one of the first reported works to propose an SSL method designed for segmenting organs in 3D US images and specifically one that incorporates unlabeled data for segmenting neonatal cerebral lateral ventricles. When compared to the state-of-the-art SSL and fully supervised learning methods, our method yielded the highest DSC and lowest VD while being computationally efficient.

PurposeDiagnostic performance of prostate MRI depends on high-quality imaging. Prostate MRI quality is inversely proportional to the amount of rectal gas and distention. Early detection of poor-quality MRI may enable intervention to remove gas or exam rescheduling, saving time. We developed a machine learning based quality prediction of yet-to-be acquired MRI images solely based on MRI rapid localizer sequence, which can be acquired in a few seconds.ApproachThe dataset consists of 213 (147 for training and 64 for testing) prostate sagittal T2-weighted (T2W) MRI localizer images and rectal content, manually labeled by an expert radiologist. Each MRI localizer contains seven two-dimensional (2D) slices of the patient, accompanied by manual segmentations of rectum for each slice. Cascaded and end-to-end deep learning models were used to predict the quality of yet-to-be T2W, DWI, and apparent diffusion coefficient (ADC) MRI images. Predictions were compared to quality scores determined by the experts using area under the receiver operator characteristic curve and intra-class correlation coefficient.ResultsIn the test set of 64 patients, optimal versus suboptimal exams occurred in 95.3% (61/64) versus 4.7% (3/64) for T2W, 90.6% (58/64) versus 9.4% (6/64) for DWI, and 89.1% (57/64) versus 10.9% (7/64) for ADC. The best performing segmentation model was 2D U-Net with ResNet-34 encoder and ImageNet weights. The best performing classifier was the radiomics based classifier.ConclusionsA radiomics based classifier applied to localizer images achieves accurate diagnosis of subsequent image quality for T2W, DWI, and ADC prostate MRI sequences.