In medical image analysis, feature engineering plays an important role in the design and performance of machine learning models. Persistent homology (PH), from the field of topological data analysis (TDA), demonstrates robustness and stability to data perturbations and addresses the limitation from traditional feature extraction approaches where a small change in input results in a large change in feature representation. Using PH, we store persistent topological and geometrical features in the form of the persistence barcode whereby large bars represent global topological features and small bars encapsulate geometrical information of the data. When multiple barcodes are computed from 2D or 3D medical images, two approaches can be used to construct the final topological feature vector in each dimension: aggregating persistence barcodes followed by featurization or concatenating topological feature vectors derived from each barcode. In this study, we conduct a comprehensive analysis across diverse medical imaging datasets to compare the effects of the two aforementioned approaches on the performance of classification models. The results of this analysis indicate that feature concatenation preserves detailed topological information from individual barcodes, yields better classification performance and is therefore a preferred approach when conducting similar experiments.

This study presents a multimodal AI framework designed for precisely classifying medical diagnostic images. Utilizing publicly available datasets, the proposed system compares the strengths of convolutional neural networks (CNNs) and different large language models (LLMs). This in-depth comparative analysis highlights key differences in diagnostic performance, execution efficiency, and environmental impacts. Model evaluation was based on accuracy, F1-score, average execution time, average energy consumption, and estimated $CO_2$ emission. The findings indicate that although CNN-based models can outperform various multimodal techniques that incorporate both images and contextual information, applying additional filtering on top of LLMs can lead to substantial performance gains. These findings highlight the transformative potential of multimodal AI systems to enhance the reliability, efficiency, and scalability of medical diagnostics in clinical settings.

PurposeTo develop and validate deep learning (DL)-based models for classifying geographic atrophy (GA) subtypes using Optical Coherence Tomography (OCT) scans across four clinical classification tasks. DesignRetrospective comparative study evaluating three DL architectures on OCT data with two experimental approaches. Subjects455 OCT volumes (258 Central GA [CGA], 74 Non-Central GA [NCGA], 123 no GA [NGA]) from 104 patients at Atrium Health Wake Forest Baptist. For GA versus age-related macular degeneration (AMD) classification, we supplemented our dataset with AMD cases from four public repositories. MethodsWe implemented ResNet50, MobileNetV2, and Vision Transformer (ViT-B/16) architectures using two approaches: (1) utilizing all B-scans within each OCT volume and (2) selectively using B-scans containing foveal regions. Models were trained using transfer learning, standardized data augmentation, and patient-level data splitting (70:15:15 ratio) for training, validation, and testing. Main Outcome MeasuresArea under the receiver operating characteristic curve (AUC-ROC), F1 score, and accuracy for each classification task (CGA vs. NCGA, CGA vs. NCGA vs. NGA, GA vs. NGA, and GA vs. other forms of AMD). ResultsViT-B/16 consistently outperformed other architectures across all classification tasks. For CGA versus NCGA classification, ViT-B/16 achieved an AUC-ROC of 0.728{+/-}0.083 and accuracy of 0.831{+/-}0.006 using selective B-scans. In GA versus NGA classification, ViT-B/16 attained an AUC-ROC of 0.950{+/-}0.002 and accuracy of 0.873{+/-}0.012 with selective B-scans. All models demonstrated exceptional performance in distinguishing GA from other AMD forms (AUC-ROC>0.998). For multi-class classification, ViT-B/16 achieved an AUC-ROC of 0.873{+/-}0.003 and accuracy of 0.751{+/-}0.002 using selective B-scans. ConclusionsOur DL approach successfully classifies GA subtypes with clinically relevant accuracy. ViT-B/16 demonstrates superior performance due to its ability to capture spatial relationships between atrophic regions and the foveal center. Focusing on B-scans containing foveal regions improved diagnostic accuracy while reducing computational requirements, better aligning with clinical practice workflows.

Dynamic MRI plays a vital role in clinical practice by capturing both spatial details and dynamic motion, but its high spatiotemporal resolution is often limited by long scan times. Deep learning (DL)-based methods have shown promising performance in accelerating dynamic MRI. However, most existing algorithms rely on large fully-sampled datasets for training, which are difficult to acquire. Recently, implicit neural representation (INR) has emerged as a powerful scan-specific paradigm for accelerated MRI, which models signals as a continuous function over spatiotemporal coordinates. Although this approach achieves efficient continuous modeling of dynamic images and robust reconstruction, it faces challenges in recovering fine details and increasing computational demands for high dimensional data representation. To enhance both efficiency and reconstruction quality, we propose TenF-INR, a novel patch-based unsupervised framework that employs INR to model bases of tensor decomposition, enabling efficient and accurate modeling of dynamic MR images with learnable tensor functions. By exploiting strong correlations in similar spatial image patches and in the temporal direction, TenF-INR enforces multidimensional low-rankness and implements patch-based reconstruction with the benefits of continuous modeling. We compare TenF-INR with state-of-the-art methods, including supervised DL methods and unsupervised approaches. Experimental results demonstrate that TenF-INR achieves high acceleration factors up to 21, outperforming all comparison methods in image quality, temporal fidelity, and quantitative metrics, even surpassing the supervised methods.

The integration of artificial intelligence (AI) with radiology signifies a transformative era in medicine. Vision foundation models have been adopted to enhance radiologic imaging analysis. However, the inherent complexities of 2D and 3D radiologic data present unique challenges that existing models, which are typically pretrained on general nonmedical images, do not adequately address. To bridge this gap and harness the diagnostic precision required in radiologic imaging, we introduce radiologic contrastive language-image pretraining (RadCLIP): a cross-modal vision-language foundational model that utilizes a vision-language pretraining (VLP) framework to improve radiologic image analysis. Building on the contrastive language-image pretraining (CLIP) approach, RadCLIP incorporates a slice pooling mechanism designed for volumetric image analysis and is pretrained using a large, diverse dataset of radiologic image-text pairs. This pretraining effectively aligns radiologic images with their corresponding text annotations, resulting in a robust vision backbone for radiologic imaging. Extensive experiments demonstrate RadCLIP's superior performance in both unimodal radiologic image classification and cross-modal image-text matching, underscoring its significant promise for enhancing diagnostic accuracy and efficiency in clinical settings. Our key contributions include curating a large dataset featuring diverse radiologic 2D/3D image-text pairs, pretraining RadCLIP as a vision-language foundation model on this dataset, developing a slice pooling adapter with an attention mechanism for integrating 2D images, and conducting comprehensive evaluations of RadCLIP on various radiologic downstream tasks.

Artificial Intelligence (AI) research needs to be clinician led; however, expertise typically lies outside their skill set. Collaborations exist but are often commercially driven. Free and open-source computational algorithms and software expertise are required for meaningful clinically driven AI medical research. Deep learning algorithms automate segmenting regions of interest for analysis and clinical translation. Numerous studies have automatically segmented paranasal sinus computed tomography (CT) scans; however, openly accessible algorithms capturing the sinonasal cavity remain scarce. The purpose of this study was to validate and provide an open-source segmentation algorithm for paranasal sinus CTs for the otolaryngology research community. A cross-sectional comparative study was conducted with a deep learning algorithm, UNet++, modified for automatic segmentation of paranasal sinuses CTs and "ground-truth" manual segmentations. A dataset of 100 paranasal sinuses scans was manually segmented, with an 80/20 training/testing split. The algorithm is available at https://github.com/rheadkaul/SinusSegment. Primary outcomes included the Dice similarity coefficient (DSC) score, Intersection over Union (IoU), Hausdorff distance (HD), sensitivity, specificity, and visual similarity grading. Twenty scans representing 7300 slices were assessed. The mean DSC was 0.87 and IoU 0.80, with HD 33.61 mm. The mean sensitivity was 83.98% and specificity 99.81%. The median visual similarity grading score was 3 (good). There were no statistically significant differences in outcomes with normal or diseased paranasal sinus CTs. Automatic segmentation of CT paranasal sinuses yields good results when compared with manual segmentation. This study provides an open-source segmentation algorithm as a foundation and gateway for more complex AI-based analysis of large datasets.

Ultrasound imaging is pivotal in clinical diagnostics due to its affordability, portability, safety, real-time capability, and non-invasive nature. It is widely utilized for examining various organs, such as the breast, thyroid, ovary, cardiac, and more. However, the manual interpretation and annotation of ultrasound images are time-consuming and prone to variability among physicians. While single-task artificial intelligence (AI) solutions have been explored, they are not ideal for scaling AI applications in medical imaging. Foundation models, although a trending solution, often struggle with real-world medical datasets due to factors such as noise, variability, and the incapability of flexibly aligning prior knowledge with task adaptation. To address these limitations, we propose an orchestration learning framework named PerceptGuide for general-purpose ultrasound classification and segmentation. Our framework incorporates a novel orchestration mechanism based on prompted hyper-perception, which adapts to the diverse inductive biases required by different ultrasound datasets. Unlike self-supervised pre-trained models, which require extensive fine-tuning, our approach leverages supervised pre-training to directly capture task-relevant features, providing a stronger foundation for multi-task and multi-organ ultrasound imaging. To support this research, we compiled a large-scale Multi-task, Multi-organ public ultrasound dataset (M²-US), featuring images from 9 organs and 16 datasets, encompassing both classification and segmentation tasks. Our approach employs four specific prompts-Object, Task, Input, and Position-to guide the model, ensuring task-specific adaptability. Additionally, a downstream synchronization training stage is introduced to fine-tune the model for new data, significantly improving generalization capabilities and enabling real-world applications. Experimental results demonstrate the robustness and versatility of our framework in handling multi-task and multi-organ ultrasound image processing, outperforming both specialist models and existing general AI solutions. Compared to specialist models, our method improves segmentation from 82.26% to 86.45%, classification from 71.30% to 79.08%, while also significantly reducing model parameters.

Accurate assessment of computed tomography (CT) image quality is crucial for ensuring diagnostic accuracy, optimizing imaging protocols, and preventing excessive radiation exposure. In clinical settings, where high-quality reference images are often unavailable, developing no-reference image quality assessment (NR-IQA) methods is essential. Recently, CT-NR-IQA methods using deep learning have been widely studied; however, significant challenges remain in handling multiple degradation factors and accurately reflecting real-world degradations. To address these issues, we propose a novel CT-NR-IQA method. Our approach utilizes a dataset that combines two degradation factors (noise and blur) to train convolutional neural network (CNN) models capable of handling multiple degradation factors. Additionally, we leveraged RadImageNet pre-trained models (ResNet50, DenseNet121, InceptionV3, and InceptionResNetV2), allowing the models to learn deep features from large-scale real clinical images, thus enhancing adaptability to real-world degradations without relying on artificially degraded images. The models' performances were evaluated by measuring the correlation between the subjective scores and predicted image quality scores for both artificially degraded and real clinical image datasets. The results demonstrated positive correlations between the subjective and predicted scores for both datasets. In particular, ResNet50 showed the best performance, with a correlation coefficient of 0.910 for the artificially degraded images and 0.831 for the real clinical images. These findings indicate that the proposed method could serve as a potential surrogate for subjective assessment in CT-NR-IQA.

Accurate patient dosimetry estimates from fluoroscopically-guided interventions (FGIs) are hindered by limited knowledge of the specific anatomy that was irradiated. Current methods use data reported by the equipment to estimate the patient anatomy exposed during each irradiation event. We propose a deep learning algorithm to automatically match 2D fluoroscopic images with corresponding anatomical regions in computational phantoms, enabling more precise patient dose estimates. Our method involves two main steps: (1) simulating 2D fluoroscopic images, and (2) developing a deep learning algorithm to predict anatomical coordinates from these images. For part (1), we utilized DeepDRR for fast and realistic simulation of 2D x-ray images from 3D computed tomography datasets. We generated a diverse set of simulated fluoroscopic images from various regions with different field sizes. In part (2), we employed a Residual Neural Network (ResNet) architecture combined with metadata processing to effectively integrate patient-specific information (age and gender) to learn the transformation between 2D images and specific anatomical coordinates in each representative phantom. For the Modified ResNet model, we defined an allowable error range of ± 10 mm. The proposed method achieved over 90% of predictions within ± 10 mm, with strong alignment between predicted and true coordinates as confirmed by Bland-Altman analysis. Most errors were within ± 2%, with outliers beyond ± 5% primarily in Z-coordinates for infant phantoms due to their limited representation in the training data. These findings highlight the model's accuracy and its potential for precise spatial localization, while emphasizing the need for improved performance in specific anatomical regions. In this work, a comprehensive simulated 2D fluoroscopy image dataset was developed, addressing the scarcity of real clinical datasets and enabling effective training of deep-learning models. The modified ResNet successfully achieved precise prediction of anatomical coordinates from the simulated fluoroscopic images, enabling the goal of more accurate patient-specific dosimetry.