Phyllodes tumors (PTs) are rare fibroepithelial breast lesions that can be malignant but are difficult to classify preoperatively due to their radiological similarity to benign fibroadenomas. This often leads to unnecessary surgical excisions. To address this, we propose a multimodal deep learning framework that integrates breast ultrasound (BUS) images with structured clinical data to improve diagnostic accuracy. We developed a dual-branch neural network that extracts and fuses features from ultrasound images and patient metadata from 81 subjects with confirmed PTs. Class-aware sampling and subject-stratified 5-fold cross-validation were applied to mitigate class imbalance and data leakage. The results show that our proposed multimodal method outperforms unimodal baselines in classifying benign versus borderline/malignant PTs. Among six deep learning-based image encoders, ConvNeXt and ResNet18 achieved the best performance in the multimodal setting, with AUC-ROC scores of 0.9427 and 0.9349, and F1-scores of 0.6720 and 0.7294, respectively. This study demonstrates the potential of multimodal AI to serve as a non-invasive diagnostic tool, reducing unnecessary surgical excisions and improving clinical decisionmaking in breast cancer care.

Early and accurate detection of Alzheimer's disease (AD) is vital for timely intervention and better patient outcomes. However, training machine learning (ML) models for this purpose is challenging due to the limited medical images available and the imbalance of classes. The size and quality of the training dataset directly affect model performance. Recent advances in diffusion models mitigate this limitation by generating synthetic images from a small sample of real images. In this work, we adapted two diffusion models and trained VGG16 and ConvNeXt classification models for AD classification. The first diffusion model was a Denoising Diffusion Probabilistic Model (DDPM) with a U-Net architecture, and the second was a U-KAN framework that integrates Kolmogorov-Arnold Networks (KANs) with the UNet. Both models were fine-tuned to generate MRI scans of AD or Late Mild Cognitive Impairment (LMCI). We conducted a comparative analysis to assess the reliability and usefulness of these synthetic images for training classification models. The best metrics achieved by the classification models using synthetic images for the AD class were precision of 96%, recall of 83%, F1 score of 87%, and AUC of 0.88. For the LMCI class, the best values were precision of 78%, recall of 88%, F1 score of 82%, and AUC of 0.88. They both demonstrated noticeable improvement from the baseline trained only on the original images.

Forecasting physiological events is essential to pre-ventive care, particularly in the context of cardiac arrhythmia, where timely intervention is crucial to saving lives. However, the complexity of physiological data has historically presented challenges in developing generalizable ML methods. In this paper, we evaluated the zero-shot performance of the recent foundation models, including Chronos, TimesFM, and Lag-Llama, on the well benchmarked MIT-BIH ECG data. Finetuning was also performed on Chronos models of varying sizes. LSTM was used as the baseline comparison. We observed that while these generative models can in some cases outperform traditional methods, the computational expense is limiting for practical applications. Our benchmarks revealed that Chronos and Lag-Llama perform more than 600 times slower than the LSTM model. However, we observe that TimesFM performs up to 179 times faster than a similarly sized Chronos model, showing that there is a substantial potential for optimization. Finetuning Chronos models yielded an average reduction of 47% in RMSE across 64 time steps, underscoring the importance of pretraining on domain relevant datasets.

Ultrasound has become an essential tool for airway management, providing non-invasive, real-time imaging that enhances the procedure efficacy. The main challenges of ultrasound include the specialized training needed for interpretation and the high patient variability. This paper presents ULTRA-AIR (Ultrasound Landmark Tracking for Real-time Anatomical Airway Identification and Reliability Check), a deep learning system designed for real-time tracking of multiple neck anatomical landmarks to assist with ultrasound-guided airway management. Uncertainty estimation was also computed to quantify model confidence scores, providing safety measures for trustworthy AI in real-time. Utilizing the YOLOv9 architecture, ULTRA-AIR was fine-tuned using Leave-One-Out cross-validation on neck ultrasound data collected from six human subjects, total of 3,060 images and 11,142 annotated bounding boxes. The system achieved AUCs of 0.99, 0.95, 0.89, and 0.77 for the classes of strap muscle, thyroid gland, tracheal ring, and thyroid cartilage, respectively. Average F1 score was 0.83 and mAP50 was 0.84. High uncertainty scores helped to identify cases needing human intervention. ULTRA-AIR shows promise for safe, accurate, and rapid airway assessments at point of care.

Automated axon tracing via fully supervised learning requires large amounts of 3D brain imagery, which is time consuming and laborious to obtain. It also requires expertise. Thus, there is a need for more efficient segmentation and centerline detection techniques to use in conjunction with automated annotation tools. Topology-preserving methods ensure that segmented components maintain geometric connectivity, which is especially meaningful for applications where volumetric data is used, and these methods often make use of morphological thinning algorithms as the thinned outputs can be useful for both segmentation and centerline detection of curvilinear structures. Current morphological thinning approaches used in conjunction with topology-preserving methods are prone to over-thinning and require manual configuration of hyperparameters.We propose an automated approach for morphological smoothing using geometric assessment of the radius of tubular structures in brain microscopy volumes, and apply average pooling to prevent over-thinning. We use this approach to formulate a loss function, which we call Geometric Assessment-driven Topological Smoothing loss, or GATS. Our approach increased segmentation and center-line detection evaluation metrics by 2%-5% across multiple datasets, and improved the Betti error rates by 9%. Our ablation study showed that geometric assessment of tubular structures achieved higher segmentation and centerline detection scores, and using average pooling for morphological smoothing in place of thinning algorithms reduced the Betti errors. We observed increased topological preservation during automated annotation of 3D axons volumes from models trained with GATS.

Many machine learning-based axon tracing methods rely on image datasets with segmentation labels. This requires manual annotation from domain experts, which is labor-intensive and not practical for large-scale brain mapping on hemisphere or whole brain tissue at cellular or sub-cellular resolution. Additionally, preserving axon structure topology is crucial to understanding neural connections and brain function. Self-supervised learning (SSL) is a machine learning framework that allows models to learn an auxiliary task on unannotated data to aid performance on a supervised target task. In this work, we propose a novel SSL auxiliary task of reconstructing an edge detector for the target task of topology-oriented axon segmentation and centerline detection. We pretrained 3D U-Nets on three different SSL tasks using a mouse brain dataset: our proposed task, predicting the order of permuted slices, and playing a Rubik’s cube. We then evaluated these U-Nets and a baseline model on a different mouse brain dataset. Across all experiments, the U-Net pretrained on our proposed task improved the baseline’s segmentation, topology-preservation, and centerline detection by up to 5.03%, 4.65%, and 5.41%, respectively. In contrast, there was no consistent improvement over the baseline observed with the slice-permutation and Rubik’s cube pretrained U-Nets.

Internal bleeding due to non-compressible torso hemorrhage is the leading cause of prehospital fatalities in civilian and military trauma. A limited number of trauma surgeons are expected to be available in disaster scenarios and future large-scale combat operations. As a result, non-specialists will need to perform life-saving interventions to address internal bleeding. A first step in mitigation is ultrasound-guided central vascular access, which involves identifying a deep blood vessel in the imagery, such as the femoral vein, femoral artery, or internal jugular vein, and then placing a needle and catheter into the vessel for follow-on resuscitation. In this paper, we demonstrate machine learning algorithms for both femoral and neck vessel detection with high accuracy and real-time speed on smartphones. The algorithms are integrated with commercial ultrasound and optimized for use on low size, weight, and power devices. Coupled with custom robotics, this technology can enable rapid vascular access by non-specialist operators using a handheld platform.

With the proliferation of wearable sensors for physiological and cognitive monitoring, a large amount of time series data needs to be processed and analyzed in a timely fashion. While deep learning has shown to be useful for the analysis, the majority of the deep learning methods are computing resource intensive. This paper demonstrates an efficient deep learning approach by adapting MINIROCKET to eye tracking and electrodermal activity data for flight performance assessment. The model was trained on 35 subjects using leave-one-subject-out cross validation and further evaluated on an independent data set of 8 subjects. We performed dimensionality reduction on each time series observation, reducing the size by 99.7% while still achieving averaged Area Under the Curve of 0.912 and average equal error rate of 0.181, thus enabling fast and accurate inference on edge devices. The approach presented here can be implemented in real-world cockpits for near instantaneous performance monitoring and could also be extended beyond this domain to other resource constrained time series applications.

This paper presents a self-supervised learning (SSL) approach for estimating the angular offset of freehand ultrasound (US) probe movement relative to an ideal viewing angle. Proper US probe angle is essential for acquiring high-quality images for diagnostics. This can be challenging, however, due to the high variability in anatomy and the subjectivity in interpretation. The ability to guide a user to an optimal probe angle can greatly improve the quality of image acquired which in turn will facilitate an accurate assessment. Probe angle is particularly important for musculoskeletal US since the probe often needs to be perpendicularly aligned with the target anatomy (e.g., bone, tendon, or ligament). In this paper, we used plantar fascia (PF) as the use case to develop a probe angle guidance pipeline. It has an immediate application for PF thickness measurement, which is used for assessing plantar fasciitis. In this SSL pipeline, circularized versions of over 3,600 US images of PF were each rotated by a varied amount, and an encoder-support vector machine pair was trained to predict this known angle of rotation. In a cross-validation from 13 subjects, the model predicted rotation angle with an average error of 2.98 degrees (SD: 2.62), and the SSL-trained model accurately predicted the relative rotation angles of images acquired while rotating the US probe. The key contribution of this work is the SSL approach on synthetically generated image and probe angle pairs to emulate angular motion relative to a standard view. This approach for estimating angular offset without labelled data has the potential to be applied to future work guiding users to an optimal view for acquiring US images.

The ability to monitor an individual's cognitive load in operational and naturalistic settings is of great importance in improving military performance and readiness. This paper describes preliminary results of using an on-body, multimodal voice monitoring system for assessing cognitive load based on vocal characteristics extracted from noise-robust sensors. The main components of the system include a commercial wired electroglottograph (EGG) and a lightweight, flex circuit that houses two sensors: an acoustic MEMS microphone (MIC) and a non-acoustic neck-surface accelerometer (ACC). We conducted human subject experiments using a cognitive load protocol under quiet laboratory conditions and computed a previously investigated vocal biomarker (creaky voice quality) from the MIC, ACC, and EGG signals. Results demonstrated the potential of discriminating low versus high cognitive load using the creaky voice correlation structure within each of the three sensor domains. Combining MIC, ACC, and EGG hardware into an integrated system would provide complementary information for quantifying voice and speech biomarkers with a high degree of robustness to competing noise sources.

In this paper, we present a lightweight, on-body, wireless system designed for monitoring real-world, ambulatory voice characteristics. The system has the potential to provide important assessments of voice and speech disorders and the impact of environmental sound levels as individuals go about their daily life. The system's transmitter is positioned on the neck and synchronously streams dual-channel sensor data from an on-board MEMS microphone and a high-bandwidth accelerometer, which acts as a noise-robust and confidential contact microphone. These data are recorded to a receiver that can store the data locally and stream a real-time feed to a computer. We also report on the design considerations of this novel system and discuss progress leading up to the latest iteration, especially of the transmitter components on a flexible circuit. Pilot data are shown from an in-field, ambulatory recording during an individual's daily activities that included settings in quiet and with naturalistic ambient noise.

A system for automatically pointing ultrasound (US) imaging catheters will enable clinicians to monitor anatomical structures and track instruments during interventional procedures. Off-the-shelf US catheters provide high quality US images from within the patient. While this method of imaging has been proven to be effective for guiding many interventional treatments, significant training is required to overcome the difficulty in manually steering the imager to point at desired structures. Our system uses closed-form four degree of freedom (DOF) kinematic solutions to automatically position the US catheter and point the imager. Algorithms for steering and imager pointing were developed for a range of useful diagnostic and interventional motions. The system was validated on a robotic test bed by steering the catheter within a water environment containing phantom objects. While the system described here was designed for pointing ultrasound catheters, these algorithms are applicable to accurate 4-DOF steering and orientation control of any long thin tendon-driven tool with single or bi-directional bending.

The common marmoset is emerging as an important transgenic model for improving the understanding of the underlying neurological basis of many brain disorders. Automated systems for quantitative monitoring of marmoset behaviors in naturalist settings over long period of time are needed to facilitate this process. This paper presents the preliminary work toward building a novel multimodal acquisition system for the automated marmoset behavior analysis in home cage. In addition to integrating commercial available devices such as Microsoft Kinect sensors and microphones of different characteristics, we also developed a wireless flexible neck collar with acoustic and non-acoustic sensors onboard for marmoset vocalization recording and caller identification. Our initial effort has been focused on the real-time synchronization of multiple sensor outputs, the engineering design of the wireless collar, and algorithms for global 3D position and local head movement from a Microsoft Kinect sensor. With limited preliminary data, we are able to estimate 3D trajectories of two marmosets with a RMSE of ~3.2 mm and track colored ear tufts with an accuracy of RMSE ~1.8 mm. A larger dataset is needed for a complete assessment and validation. Our system architecture is modular and flexible, and can be extended to include more sensors and devices if needed.

Intracardiac echocardiography (ICE) catheters enable high-quality ultrasound imaging within the heart, but their use in guiding procedures is limited due to the difficulty of manually pointing them at structures of interest. This paper presents the design and testing of a catheter steering model for robotic control of commercial ICE catheters. The four actuated degrees of freedom (4-DOF) are two catheter handle knobs to produce bi-directional bending in combination with rotation and translation of the handle. An extra degree of freedom in the system allows the imaging plane (dependent on orientation) to be directed at an object of interest. A closed form solution for forward and inverse kinematics enables control of the catheter tip position and the imaging plane orientation. The proposed algorithms were validated with a robotic test bed using electromagnetic sensor tracking of the catheter tip. The ability to automatically acquire imaging targets in the heart may improve the efficiency and effectiveness of intracardiac catheter interventions by allowing visualization of soft tissue structures that are not visible using standard fluoroscopic guidance. Although the system has been developed and tested for manipulating ICE catheters, the methods described here are applicable to any long thin tendon-driven tool (with single or bi-directional bending) requiring accurate tip position and orientation control.