Radiologists currently have very limited and time-consuming options to annotate findings on the images and are mostly limited to arrows, calipers and lines to annotate any type of findings on most PACS systems. We propose a framework placing encoded, transferable, highly contextual structured text annotations directly on PACS images indicating the type of lesion, level of suspicion, location, lesion measurement, and TNM status for malignant lesions, along with automated integration of this information into the radiology report. This approach offers a one-stop solution to generate radiology reports that are easily understood by other radiologists, patient care providers, patients, and machines while reducing the effort needed to dictate a detailed radiology report and minimizing speech recognition errors. It also provides a framework for automated generation of large volume high quality annotated data sets for machine learning algorithms from daily work of radiologists. Enabling voice dictation of these contextual annotations directly into PACS similar to voice enabled Google search will further enhance the user experience. Wider adaptation of contextualized structured annotations in the future can facilitate studies understanding the temporal evolution of different tumor lesions across multiple lines of treatment and early detection of asynchronous response/areas of treatment failure. We present a futuristic vision, and solution with the potential to transform clinical work and research in oncologic imaging.

An effective diagnosis system and suitable treatment planning require the precise segmentation of thyroid nodules in ultrasound imaging. The advancement of imaging technologies has not resolved traditional imaging challenges, which include noise issues, limited contrast, and dependency on operator choices, thus highlighting the need for automated, reliable solutions. The researchers developed TATHA, an innovative deep learning architecture dedicated to improving thyroid ultrasound image segmentation accuracy. The model is evaluated using the digital database of thyroid ultrasound images, which includes 99 cases across three subsets containing 134 labelled images for training, validation, and testing. It incorporates data pre-treatment procedures that reduce speckle noise and enhance contrast, while edge detection provides high-quality input for segmentation. TATHA outperforms U-Net, PSPNet, and Vision Transformers across various datasets and cross-validation folds, achieving superior Dice scores, accuracy, and AUC results. The distributed thyroid segmentation framework generates reliable predictions by combining results from multiple feature extraction units. The findings confirm that these advancements make TATHA an essential tool for clinicians and researchers in thyroid imaging and clinical applications.

Ulcerative colitis (UC) is a chronic inflammatory disorder necessitating precise severity stratification to facilitate optimal therapeutic interventions. This study harnesses a triple-pronged deep learning methodology-including multimodal inference pipelines that eliminate domain-specific training, few-shot meta-learning, and Vision Transformer (ViT)-based ensembling-to classify UC severity within the HyperKvasir dataset. We systematically evaluate multiple vision transformer architectures, discovering that a Swin-Base model achieves an accuracy of 90%, while a soft-voting ensemble of diverse ViT backbones boosts performance to 93%. In parallel, we leverage multimodal pre-trained frameworks (e.g., CLIP, BLIP, FLAVA) integrated with conventional machine learning algorithms, yielding an accuracy of 83%. To address limited annotated data, we deploy few-shot meta-learning approaches (e.g., Matching Networks), attaining 83% accuracy in a 5-shot context. Furthermore, interpretability is enhanced via SHapley Additive exPlanations (SHAP), which interpret both local and global model behaviors, thereby fostering clinical trust in the model's inferences. These findings underscore the potential of contemporary representation learning and ensemble strategies for robust UC severity classification, highlighting the pivotal role of model transparency in facilitating medical image analysis.

Diffusion MRI (dMRI) is an advanced imaging technique that enables in-vivo tracking of white matter fiber tracts and estimates the underlying cellular microstructure of brain tissues. Groupwise registration of dMRI data from multiple individuals is an important task for brain template creation and investigation of inter-subject brain variability. However, groupwise registration is a challenging task due to the uniqueness of dMRI data that include multi-dimensional, orientation-dependent signals that describe not only the strength but also the orientation of water diffusion in brain tissues. Deep learning approaches have shown successful performance in standard subject-to-subject dMRI registration. However, no deep learning methods have yet been proposed for groupwise dMRI registration. . In this work, we propose Deep Diffusion MRI Template (DDTemplate), which is a novel deep-learning-based method building upon the popular VoxelMorph framework to take into account dMRI fiber tract information. DDTemplate enables joint usage of whole-brain tissue microstructure and tract-specific fiber orientation information to ensure alignment of white matter fiber tracts and whole brain anatomical structures. We propose a novel deep learning framework that simultaneously trains a groupwise dMRI registration network and generates a population brain template. During inference, the trained model can be applied to register unseen subjects to the learned template. We compare DDTemplate with several state-of-the-art registration methods and demonstrate superior performance on dMRI data from multiple cohorts (adolescents, young adults, and elderly adults) acquired from different scanners. Furthermore, as a testbed task, we perform a between-population analysis to investigate sex differences in the brain, using the popular Tract-Based Spatial Statistics (TBSS) method that relies on groupwise dMRI registration. We find that using DDTemplate can increase the sensitivity in population difference detection, showing the potential of our method's utility in real neuroscientific applications.

Artificial Intelligence (AI), and particularly deep learning (DL), has shown great promise to revolutionize healthcare. However, clinical translation is often hindered by demanding hardware requirements. In this study, we assess the effectiveness of optimization techniques for DL models in healthcare applications, targeting varying AI workloads across the domains of radiology, histopathology, and medical RGB imaging, while evaluating across hardware configurations. The assessed AI workloads focus on both segmentation and classification workloads, by virtue of brain extraction in Magnetic Resonance Imaging (MRI), colorectal cancer delineation in Hematoxylin & Eosin (H&E) stained digitized tissue sections, and diabetic foot ulcer classification in RGB images. We quantitatively evaluate model performance in terms of model runtime during inference (including speedup, latency, and memory usage) and model utility on unseen data. Our results demonstrate that optimization techniques can substantially improve model runtime, without compromising model utility. These findings suggest that optimization techniques can facilitate the clinical translation of AI models in low-resource environments, making them more practical for real-world healthcare applications even in underserved regions.

Recent advances in deep learning have significantly enhanced the performance of medical image segmentation. However, maintaining a balanced integration of feature localization, global context modeling, and computational efficiency remains a critical research challenge. Convolutional Neural Networks (CNNs) effectively capture fine-grained local features through hierarchical convolutions; however, they often struggle to model long-range dependencies due to their limited receptive field. Transformers address this limitation by leveraging self-attention mechanisms to capture global context, but they are computationally intensive and require large-scale data for effective training. The Mamba architecture has emerged as a promising approach, effectively capturing long-range dependencies while maintaining low computational overhead and high segmentation accuracy. Based on this, we propose a method named CLT-MambaSeg that integrates Convolution, Linear Transformer, and Multiscale Mamba architectures to capture local features, model global context, and improve computational efficiency for medical image segmentation. It utilizes a convolution-based Spatial Representation Extraction (SREx) module to capture intricate spatial relationships and dependencies. Further, it comprises a Mamba Vision Linear Transformer (MVLTrans) module to capture multiscale context, spatial and sequential dependencies, and enhanced global context. In addition, to address the problem of limited data, we propose a novel Memory-Guided Augmentation Generative Adversarial Network (MeGA-GAN) that generates synthetic realistic images to further enhance the segmentation performance. We conduct extensive experiments and ablation studies on the five benchmark datasets, namely CVC-ClinicDB, Breast UltraSound Images (BUSI), PH2, and two datasets from the International Skin Imaging Collaboration (ISIC), namely ISIC-2016 and ISIC-2017. Experimental results demonstrate the efficacy of the proposed CLT-MambaSeg compared to other state-of-the-art methods.

Computed tomography (CT) images are reconstructed from raw datasets including sinogram using various convolution kernels through back projection. Kernels are typically chosen depending on the anatomical structure being imaged and the specific purpose of the scan, balancing the trade-off between image sharpness and pixel noise. Generally, a sinogram requires large storage capacity, and storage space is often limited in clinical settings. Thus, CT images are generally reconstructed with only one specific kernel in clinical settings, and the sinogram is typically discarded after a week. Therefore, many researchers have proposed deep learning-based image-to-image translation methods for CT kernel conversion. However, transferring the style of the target kernel while preserving anatomical structure remains challenging, particularly when translating CT images from a source domain to a target domain in an unpaired manner, which is often encountered in real-world settings. Thus, we propose a novel kernel conversion method using unpaired image-to-image translation (KC-UNIT). This approach utilizes discriminator regularization, using feature maps from the generator to improve semantic representation learning. To capture content and style features, cosine similarity content and contrastive style losses were defined between the feature map of generator and semantic label map of discriminator. This can be easily incorporated by modifying the discriminator's architecture without requiring any additional learnable or pre-trained networks. The KC-UNIT demonstrated the ability to preserve fine-grained anatomical structure from the source domain during transfer. Our method outperformed existing generative adversarial network-based methods across most kernel conversion methods in three kernel domains. The code is available at https://github.com/cychoi97/KC-UNIT.

A key cardiac magnetic resonance (CMR) challenge is breath-holding duration, difficult for cardiac patients. To evaluate whether artificial intelligence-assisted compressed sensing CINE (AI-CS-CINE) reduces image acquisition time of CMR compared to conventional CINE (C-CINE). Cardio-oncology patients (<i>n</i> = 60) and healthy volunteers (<i>n</i> = 29) underwent sequential C-CINE and AI-CS-CINE with a 1.5-T scanner. Acquisition time, visual image quality assessment, and biventricular metrics (end-diastolic volume, end-systolic volume, stroke volume, ejection fraction, left ventricular mass, and wall thickness) were analyzed and compared between C-CINE and AI-CS-CINE with Bland-Altman analysis, and calculation of intraclass coefficient (ICC). In 89 participants (58.5 ± 16.8 years, 42 males, 47 females), total AI-CS-CINE acquisition and reconstruction time (37 seconds) was 84% faster than C-CINE (238 seconds). C-CINE required repeats in 23% (20/89) of cases (approximately 8 minutes lost), while AI-CS-CINE only needed one repeat (1%; 2 seconds lost). AI-CS-CINE had slightly lower contrast but preserved structural clarity. Bland-Altman plots and ICC (0.73 ≤ <i>r</i> ≤ 0.98) showed strong agreement for left ventricle (LV) and right ventricle (RV) metrics, including those in the cardiac amyloidosis subgroup (<i>n</i> = 31). AI-CS-CINE enabled faster, easier imaging in patients with claustrophobia, dyspnea, arrhythmias, or restlessness. Motion-artifacted C-CINE images were reliably interpreted from AI-CS-CINE. AI-CS-CINE accelerated CMR image acquisition and reconstruction, preserved anatomical detail, and diminished impact of patient-related motion. Quantitative AI-CS-CINE metrics agreed closely with C-CINE in cardio-oncology patients, including the cardiac amyloidosis cohort, as well as healthy volunteers regardless of left and right ventricular size and function. AI-CS-CINE significantly enhanced CMR workflow, particularly in challenging cases. The strong analytical concordance underscores reliability and robustness of AI-CS-CINE as a valuable tool.

Anomaly detection in medical imaging plays a crucial role in identifying pathological regions across various imaging modalities, such as brain MRI, liver CT, and carotid ultrasound (US). However, training fully supervised segmentation models is often hindered by the scarcity of expert annotations and the complexity of diverse anatomical structures. To address these issues, we propose a novel unsupervised anomaly detection framework based on a diffusion model that incorporates a synthetic anomaly (Synomaly) noise function and a multi-stage diffusion process. Synomaly noise introduces synthetic anomalies into healthy images during training, allowing the model to effectively learn anomaly removal. The multi-stage diffusion process is introduced to progressively denoise images, preserving fine details while improving the quality of anomaly-free reconstructions. The generated high-fidelity counterfactual healthy images can further enhance the interpretability of the segmentation models, as well as provide a reliable baseline for evaluating the extent of anomalies and supporting clinical decision-making. Notably, the unsupervised anomaly detection model is trained purely on healthy images, eliminating the need for anomalous training samples and pixel-level annotations. We validate the proposed approach on brain MRI, liver CT datasets, and carotid US. The experimental results demonstrate that the proposed framework outperforms existing state-of-the-art unsupervised anomaly detection methods, achieving performance comparable to fully supervised segmentation models in the US dataset. Ablation studies further highlight the contributions of Synomaly noise and the multi-stage diffusion process in improving anomaly segmentation. These findings underscore the potential of our approach as a robust and annotation-efficient alternative for medical anomaly detection. Code:https://github.com/yuan-12138/Synomaly.

Fast radial-MRI approaches based on compressed sensing (CS) and deep learning (DL) often use non-uniform fast Fourier transform (NUFFT) as the forward imaging operator, which might introduce interpolation errors and reduce image quality. Using the polar Fourier transform (PFT), we developed fully polar CS and DL algorithms for fast 2D cardiac radial-MRI. Our methods directly reconstruct images in polar spatial space from polar k-space data, eliminating frequency interpolation and ensuring an easy-to-compute data consistency term for the DL framework via the variable splitting (VS) scheme. Furthermore, PFT reconstruction produces initial images with fewer artifacts in a reduced field of view, making it a better starting point for CS and DL algorithms, especially for dynamic imaging, where information from a small region of interest is critical, as opposed to NUFFT, which often results in global streaking artifacts. In the cardiac region, PFT-based CS technique outperformed NUFFT-based CS at acceleration rates of 5x (mean SSIM: 0.8831 vs. 0.8526), 10x (0.8195 vs. 0.7981), and 15x (0.7720 vs. 0.7503). Our PFT(VS)-DL technique outperformed the NUFFT(GD)-based DL method, which used unrolled gradient descent with the NUFFT as the forward imaging operator, with mean SSIM scores of 0.8914 versus 0.8617 at 10x and 0.8470 versus 0.8301 at 15x. Radiological assessments revealed that PFT(VS)-based DL scored 2.9±0.30 and 2.73±0.45 at 5x and 10x, whereas NUFFT(GD)-based DL scored 2.7±0.47 and 2.40±0.50, respectively. Our methods suggest a promising alternative to NUFFT-based fast radial-MRI for dynamic imaging, prioritizing reconstruction quality in a small region of interest over whole image quality.