An advanced AI-driven system for synthetic medical data generation and precise segmentation of cardiac MRI to enhance accuracy and efficiency in cardiovascular health.

Researchers at the University of California, Davis have developed an AI-powered framework that bridges the gap between predictive feature analysis and actionable interventions by extracting evidence-based recommendations from scientific literature.

Historically, speech therapy apps have relied primarily on online cloud-based speech recognition systems like those used in digital assistants (Cortana, Siri, Google Assistant), which are designed to best guess speech rather than critically evaluate articulation errors. For children with cleft palate specifically, affecting 1 in 700 babies globally, speech therapy is essential follow-up care after reconstructive surgery. Approximately 25% of children with clefts use compensatory articulation errors, and when these patterns become habituated during ages 3-5, they become particularly resistant to change in therapy. Traditional approaches to mobile speech therapy apps have included storybook-style narratives that proved expensive with low replayability and engagement, as well as fast-paced arcade-style games that failed to maintain user interest. Common speech therapy applications require a facilitator to evaluate speech performance and typically depend on continuous internet connectivity, creating barriers for users in areas with poor network coverage or those concerned about data privacy and roaming costs. The shift toward gamified therapy solutions showed that game elements can serve as powerful motivators for otherwise tedious activities. Speech recognition systems face inherent limitations in accuracy compared to cloud-based solutions and require substantial processing power and memory that can impact device performance and battery life, particularly on older mobile devices. Automatic speech recognition (ASR) models struggle significantly with children's speech due to non-fluent pronunciation and variability in speech patterns, with phoneme error rates reaching almost 12%, and consonant recognition errors affecting the reliability of speech disorder detection. The challenge becomes even more pronounced for populations with speech impairments, as conventional ASR systems are optimized for typical adult speech rather than atypical articulation patterns of cleft palate speech or developmental disabilities. Moreover, maintaining user engagement over extended therapy periods is hard, and many apps fail to provide sufficient motivation for daily practice, which is essential for speech improvement.

A medical device that uses machine learning and augmented reality to project precise surgical guides onto 3D patient anatomy, enabling real-time surgical guidance and remote expert collaboration.

QAFeL is a novel asynchronous federated learning framework that combines buffered aggregation with bidirectional quantized communications, achieving up to 8× lower communication costs while preserving convergence speed and accuracy.

A groundbreaking algorithm that significantly reduces communication time and message size in distributed machine learning, ensuring fast and reliable model convergence.

An advanced geometric method for comprehensive 3D cardiac strain analysis, enhancing diagnosis and monitoring of myocardial diseases.

Advances in biological research have been greatly influenced by the development of organoids, a specialized form of 3D cell culture. Created from pluripotent stem cells, organoids are effective in vitro models in replicating the structure and progression of organ development, providing an exceptional tool for studying the complexities of biology. Among these, cerebral cortex organoids (hereafter "organoid") have become particularly instrumental in providing valuable insights into brain formation, function, and pathology. Modern methods of interfacing with organoids involve any combination of encoding information, decoding information, or perturbing the underlying dynamics through various timescales of plasticity. Our knowledge of biological learning rules has not yet translated to reliable methods for consistently training neural tissue in goal-directed ways. In vivo training methods commonly exploit principles of reinforcement learning and Hebbian learning to modify biological networks. However, in vitro training has not seen comparable success, and often cannot utilize the underlying, multi-regional circuits enabling dopaminergic learning. Successfully harnessing in vitro learning methods and systems could uniquely reveal fundamental mesoscale processing and learning principles. This may have profound implications, from developing targeted stimulation protocols for therapeutic interventions to creating energy-efficient bio-electronic systems.

A revolutionary software that integrates with surgical microscopes to accurately locate the trabecular meshwork (TM), enhancing the safety and efficiency of glaucoma surgeries.

In PET imaging, patient motion, such as respiratory and cardiac motion, are a major source of blurring and motion artifacts. Researchers at the University of California, Davis have developed a technology designed to enhance PET imaging resolution without the need for external devices by effectively mitigating these artifacts

A novel method using Diffusion Tensor Imaging (DTI) combined with Statistical Parametric Mapping (SPM) as an effective diagnostic tool for Traumatic Brain Injury.

An innovative system designed to enhance rehabilitation therapy for neurological conditions through comprehensive, computer-based solutions.

An AI-powered platform for the generation of automated electronic patient anatomy education models, providing surgeons with clinically relevant patient anatomy data.

An innovative algorithm linking electroencephalogram (EEG) neural data with cognitive model parameters to predict brain signals from behavioral data.

      Poorly designed healthcare environments can increase patient stress and delay recovery, particularly in pediatric settings (see, e.g., Devlin & Andrade 2017; Park et al. 2018; Jafarifiroozabadi et al. 2023). Traditional methods for gathering architectural design feedback, such as interviews, surveys, and focus groups, rely heavily on subjective user input, and often fail to capture the voices of children by relying on parent proxies. Physical mock-ups, a common alternative to traditional methods, provide a full-scale model of a room or space, often constructed from materials like cardboard or foam. While these mock-ups allow for some degree of spatial exploration, they are time-intensive, and limited in their ability to replicate real-world conditions; high-fidelity mock-ups which incorporate more realistic materials and finishes add expense and limit flexibility for testing multiple design iterations.       To overcome these challenges UC Berkeley researchers have developed an innovative participatory design methodology that leverages advanced virtual reality (VR), eye-tracking, and physiological/emotional biofeedback technologies to evaluate the design of pediatric healthcare environments. This comprehensive system is further enhanced by custom-developed workflows for creating dynamic, interactive room simulations that are randomized to ensure rigorous, unbiased data collection. The methodology is uniquely capable of gathering objective, quantifiable data on how pediatric patients and their families respond physiologically and emotionally to specific environmental design features.

Mental disorders like depression and anxiety are often hard to diagnose and evaluate because there aren't any objective measures for them. This makes it difficult to tell the difference between these disorders and other conditions with similar symptoms. This invention, developed by UC Berkeley researchers, addresses this challenge by providing methods, compositions, and systems that use functional biomarkers to diagnose and evaluate mental disorders. The technology combines brain imaging techniques with a brain encoding model to objectively identify and assess these conditions. This approach offers a more precise and data-driven way to diagnose and evaluate mental disorders compared to traditional, subjective methods.

Researchers at the University of California, Davis have developed a novel cell segmentation technology for accurate analysis of non-spherical cells and that offers a comprehensive, high-throughput approach for analyzing the transcriptomic and metabolomic data to study complex biological processes at the single-cell level.

Brief description not available

Deoxyribonucleic acid- (DNA-) based identification in forensics is typically accomplished via genotyping allele length at a defined set of short tandem repeat (STR) loci via polymerase chain reaction (PCR). These PCR assays are robust, reliable, and inexpensive. Given the multiallelic nature of each of these loci, a small panel of STR markers can provide suitable discriminatory power for personal identification. Massively parallel sequencing (MPS) technologies and genotype array technologies invite new approaches for DNA-based identification. Application of these technologies has provided catalogs of global human genetic variation at single-nucleotide polymorphic (SNP) sites and short insertion-deletion (INDEL) sites. For example, from the 1000 Genomes Project, there is now a catalog of nearly all human SNP and INDEL variation down to 1% worldwide frequency. Genotype files, generated via MPS or genotype array, can be compared between individuals to find regions that are co-inherited or identical-by-descent (IBD). These comparisons are the basis of the relative finder functions in many direct-to-consumer genetic testing products. A special case of relative-finding is self-identification. This is a trivial comparison of genotype files as self-comparisons will be identical across all sites, minus the error rate of the assay. For many forensic samples, however, the available DNA may not be suitable for PCR-based STR amplification, genotype array analysis, or MPS to the depth required for comprehensive, accurate genotype calling. In the case of PCR, one of the most common failure modes occurs when DNA is too fragmented for amplification. For these samples, it may be possible to directly observe the degree of DNA fragmentation from the decreased amplification efficiency of larger STR amplicons from a multiplex STR amplification. In the case of severely fragmented samples, where all DNA fragments are shorter than the shortest STR amplicon length, PCR simply fails with no product.

Diagnosing retinal disease, which affects over 200 million people worldwide, requires expensive and complicated analysis of the structure and function of retinal tissue. Recently, UCI developed a training algorithm which, for the first time, is able to assess tissue health from images collected using more common and less expensive optics.

Dysfluent speech modeling requires time-accurate and silence-aware transcription at both the word-level and phonetic-level. However, current research in dysfluency modeling primarily focuses on either transcription or detection, and the performance of each aspect remains limited.To address this problem, UC Berkeley researchers have developed a new unconstrained dysfluency modeling (UDM) approach that addresses both transcription and detection in an automatic and hierarchical manner. Furthermore, a simulated dysfluent dataset called VCTK++ enhances the capabilities of UDM in phonetic transcription. The effectiveness and robustness of UDM in both transcription and detection tasks has been demonstrated experimentally.UDM eliminates the need for extensive manual annotation by providing a comprehensive solution.

Electrical properties (EP), such as permittivity and conductivity, dictate the interactions between electromagnetic waves and biological tissue. EP are biomarkers for pathology characterization, such as cancer. Imaging of EP helps monitor the health of the tissue and can provide important information in therapeutic procedures. Magnetic resonance (MR)-based electrical properties tomography (MR-EPT) uses MR measurements, such as the magnetic transmit field B1+, to reconstruct EP. These reconstructions rely on the calculations of spatial derivatives of the measured B1+. However, the numerical approximation of derivatives leads to noise amplifications introducing errors and artifacts in the reconstructions. Recently, a supervised learning-based method (DL-EPT) has been introduced to reconstruct robust EP maps from noisy measurements. Still, the pattern-matching nature of this method does not allow it to generalize for new samples since the network’s training is done on a limited number of simulated data pairs, which makes it unrealistic in clinical applications. Thus, there is a need for a robust and realistic method for EP map construction.

Researchers at UC Irvine have developed a novel algorithm that more accurately filters raw blood pressure (BP) data collected from continuous non-invasive blood pressure sensors. The algorithm features improvements in eliminating baseline signal drift while maintaining signal integrity and BP estimation accuracy across significant hemodynamic changes.

Common nuclear imaging techniques include computed tomography (CT), single photon emission CT (SPECT), and positron emission tomography (PET). PET differs from other nuclear imaging techniques in that it can visualize both functional and biological activities, including detection of metabolism within human tissues. PET is especially good for imaging patients with cancer, or brain or heart conditions. At low energies, when positrons collide with electrons near the radionuclide decay, Gamma rays (annihilation photons) are created. Gammas originating from the same electron-positron annihilation are generated exclusively in an entangled Bell state. Gammas which do not share an annihilation origin event, such as randoms, are not entangled. Additionally, a gamma which undergoes an internal scatter becomes decoherent (unentangled) from its pair, such as the gammas found in the scattered coincidence pairs. Scattered and random events degrade the image quality. Recently, quantum-based techniques utilizing entanglement of annihilation photons has been recognized as one approach to address scatter and random and to optimize the signal to noise (SNR) ratio.

Measurement of electrical activity in nervous tissue has many applications in medicine, but the implantation of a large number of sensors is traditionally very risky and costly. Devices must be large due to their necessary complexity and power requirements, driving up the risk further and discouraging adoption. To address these problems, researchers at UC Berkeley have developed devices and methods to allow small, very simple and power-efficient sensors to transmit information by backscatter feedback. That is, a much more complex and powerful external interrogator sends an electromagnetic or ultrasound signal, which is modulated by the sensor nodes and reflected back to the interrogator. Machine learning algorithms are then able to map the reflected signals to nervous activity. The asymmetric nature of this process allows most of the complexity to be offloaded to the external interrogator, which is not subject to the same constraints as implanted devices. This allows for larger networks of nodes which can generate higher resolution data at lower risks and costs than existing devices.