Radically different
Nature Medicine, Published online: 19 May 2026; doi:10.1038/s41591-026-04398-3
In a prospective cohort of 166 participants (and in a 63-participant validation cohort), dermal α-synuclein and 4-repeat tau seed amplification, combined with serum neurofilament light chain assays, were able to identify parkinsonian syndromes.
Researchers at the University of California, San Francisco (UCSF) have developed an artificial intelligence model capable of predicting cognitive impairment and Alzheimer’s disease progression using only a single baseline MRI scan and basic demographic information. The approach, published in Nature Aging, could help make early Alzheimer’s assessment faster, more accessible, and less dependent on costly specialized testing.
Alzheimer’s disease accounts for approximately 60% to 70% of dementia cases worldwide. Although structural brain changes and cognitive decline are hallmarks of the disease, accurately forecasting who will develop progressive impairment remains difficult.
Current diagnostic workflows often rely on multiple complementary techniques, including PET imaging, cerebrospinal fluid or blood biomarkers, genetic testing, and comprehensive neuropsychological assessments. While effective, these approaches can be expensive, time-consuming, and inaccessible in many healthcare settings.
MRI scans are among the most widely available clinical imaging tools for neurological assessment, but MRI data alone has historically struggled to capture the complexity and heterogeneity of Alzheimer’s disease progression when used in conventional AI frameworks.
To address this challenge, the UCSF team designed a multitask deep learning framework that combines domain-specific imaging knowledge with advanced machine learning methods to predict cognitive outcomes directly from structural MRI scans.
Unlike many earlier Alzheimer’s prediction models, the new system does not require longitudinal imaging data, baseline cognitive testing, PET scans, or molecular biomarker analysis.
The researchers instead focused on extracting clinically meaningful information from a single baseline MRI scan. The framework was trained to perform several related tasks simultaneously, including tissue segmentation, Alzheimer’s diagnosis prediction, and estimation of both present and future cognitive performance.
A key innovation of the study was the development of a specialized image model that segments brain tissue into gray matter, white matter, and cerebrospinal fluid before generating cognitive predictions. According to the authors, this task-specific segmentation step allowed the model to learn biologically relevant spatial brain features more effectively than standard transfer-learning approaches.
Senior study author Ashish Raj, PhD, professor of radiology and biomedical imaging at UCSF, said the goal was to create a system that could be realistically implemented in routine clinical environments.
“Unlike previous approaches, our model does not require baseline cognitive assessment, specialized image pipelines, expensive PET scans, genetic analysis, or fluid proteomics, making it a fast, accurate, and easily implementable tool for most clinical settings,” Raj said in a statement.
To train and validate the framework, the researchers used imaging and clinical data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), including MRI scans, demographic information, diagnoses, and cognitive assessments.
The team also incorporated MRI data from the Human Connectome Project Young Adult cohort, which contains scans from healthy younger adults with minimal age-related brain atrophy. According to the authors, exposing the model to healthy brain anatomy improved its ability to distinguish pathological neurodegeneration from normal aging.
An external validation cohort from the Dallas Lifespan Brain Study was additionally used to test the generalizability of the framework across independent datasets.
The researchers reported that the multitask framework outperformed existing AI methods, including standard transfer-learning approaches, in predicting clinically relevant outcomes. The model generated accurate predictions for Alzheimer’s diagnosis, tissue segmentation, current cognitive function, and future cognitive decline using only baseline MRI data.
The study also reported improvements in computational efficiency and processing speed compared with more complex MRI morphometry pipelines commonly used in neuroimaging research.
First author Daren Ma, MSc, a machine learning specialist in the Raj Lab at UCSF, said the framework could help clinicians identify at-risk patients earlier and streamline referrals for advanced neurological evaluation.
“We reported meaningful gains in speed and performance over other pipelines, which could prove valuable in developing a quick clinical prediction of cognitive impairment prior to referring the patient to a more advanced imaging lab and/or a full neuroradiology report,” Ma said.
The researchers believe the framework could eventually be adapted for other neurodegenerative disorders characterized by structural brain changes and progressive cognitive decline.
Potential future applications include Parkinson’s disease, amyotrophic lateral sclerosis (ALS), and Huntington’s disease. The ability to estimate cognitive impairment using minimal baseline data may also prove useful in community healthcare settings where access to specialist neuropsychological testing is limited.
In addition, the model may have implications for clinical trial design. Identifying likely disease progressors early could help reduce trial size requirements and improve patient selection for studies evaluating disease-modifying therapies.
“The ability to correctly predict progressors from non-progressors using only baseline data can dramatically reduce sample sizes and cost,” Raj said.
The authors emphasized, however, that further validation will be necessary before the model can be broadly implemented in routine clinical practice. Future iterations of the framework may incorporate additional clinical measurements where available, including longitudinal MRI imaging, PET scans, genetics, and blood or cerebrospinal fluid biomarkers.
The study highlights the growing role of AI-driven imaging analysis in neurology and suggests that clinically accessible tools such as MRI may eventually support earlier and more scalable prediction of Alzheimer’s disease progression.
The post AI Model Predicts Alzheimer’s Progression from a Single MRI Scan appeared first on Inside Precision Medicine.
A research team at the University of California, San Diego has developed a machine learning-based screening approach for peripheral artery disease (PAD) that uses a light-based technology called photoplethysmography (PPG) that can measure changes in blood volume in tissue. The researchers reported that short-duration PPG recordings in a patient’s toe, analyzed by machine learning models, identified PAD with a high degree of accuracy and may provide the basis for a scalable digital screening tool that could eventually be deployed through smartphones, pulse oximeters, and wearable devices. The team’s findings are published in npj Digital Medicine.
“PPG works by shining a light into tissue, in our case, the toe,” said co-first author Ava J. Fascetti, a PhD student in the digital health technology lab of senior author Edward J. Wang, PhD. “A photosensor measures how much light is reflected back, allowing us to detect tiny changes in blood volume: what we call the PPG signal.”
PAD is caused by plaque buildup in arteries that restricts blood flow, particularly to the legs and lower extremities. The disease affects an estimated 12 million Americans and 200 million adults worldwide. PAD substantially increases the risk of limb loss and major cardiovascular events, yet many patients are not diagnosed until later stages of disease progression. The researchers noted that the condition disproportionately affects underserved populations and is underdiagnosed in part because the current standard diagnostic, ankle-brachial index (ABI), requires specialized equipment, staff, and clinic visits.
“There exists a glaring unmet clinical need to develop technology to meet the demands of modern practice,” the researchers wrote. Further, ABI testing, introduced about 60 years ago, has has remained largely unchanged and has long-standing barriers to widespread use in primary care and under-resourced settings.
The current study originated from discussions between co-first author Mattheus Ramsis, MD, and assistant professor of medicine and medical director of cardiology informatics, and co-author Elsie G. Ross, MD, an associate professor of surgery in vascular and endovascular surgery, who noted that vascular labs conducting ABI testing often also collected toe PPG waveforms.
“The light-bulb went off for me at that moment,” Ramsis said.
PPG works by shining light into tissue and measuring backscattered light associated with blood volume changes. PPG has previously been used to identify cardiovascular and metabolic conditions including diabetes and atrial fibrillation. Earlier research efforts to use PPG for PAD detection had relied on small datasets, long recordings and less interpretable deep-learning approaches.
For their approach, the UCSD team assembled a dataset containing more than 10,000 toe PPG recordings from more than 3,500 patients who underwent ABI testing at UC San Diego Health between 2020 and 2025. Using these data, the researchers extracted 78 waveform features from the PPG signals that correlated significantly with ABI measurements. Those features were then used to train an explainable support vector machine model designed to identify PAD from PPG data alone.
Ramsis said the model correctly distinguished PAD cases approximately 83% of the time using only PPG data, compared with roughly 60% to 65% performance typically achieved using clinical risk-factor assessments alone. Incorporating smoking status of the patients further improved the performance of the new method.
Importantly, the model performed consistently across Black, Hispanic, and White patient populations, and among patients with diabetes, coronary artery disease, and end-stage renal disease. The researchers also reported similar performance across two UC San Diego Health campuses that used different equipment and staff.
The investigators noted that the physiologic basis for their findings align with established vascular biology. In PAD, reduced blood flow and arterial stiffness alter the morphology of PPG waveforms. Healthier patients demonstrated steeper systolic upstrokes and narrower waveform widths, while patients with PAD showed more dampened signals.
“Our findings support the existence of a reproducible PPG-derived digital biomarker that captures peripheral vascular pathophysiology relevant to ABI-defined PAD,” the researchers wrote.
The researchers said they don’t think their new model should replace ABI testing. Instead, they envision PPG screening as a complementary tool that could serve to identify patients earlier that might need further vascular evaluation.
The team said prospective deployment studies are already underway to evaluate performance in clinical settings and across additional reference standards, including toe pressure measurements, ultrasound imaging, and angiography. Additional research will also gauge performance in consumer-grade environments, including smartphones and wearable devices, and assess how the screening tool functions in broader patient populations outside specialized vascular clinics.
“If we can catch PAD early enough to prevent a limb amputation, that would be the ultimate impact: preserving limb function, reducing mortality, and addressing barriers in underserved populations,” Ramsis said.
The post PPG-Derived Digital Biomarker Developed for Peripheral Artery Disease Detection appeared first on Inside Precision Medicine.
Ovarian cancer remains one of the deadliest gynaecological malignancies, with high-grade serous ovarian cancer (HGSOC) accounting for the majority of deaths.1,2 Despite advances in surgery and chemotherapy, most patients are diagnosed at an advanced stage, and the 5-year survival rate has stubbornly remained below 50% for decades.3 A major obstacle is that the traditional FIGO staging system, while useful for prognosis, does not explain why patients within the same stage often follow dramatically different clinical trajectories.
Scientists have developed a sensor platform that can monitor cerebrospinal fluid (CSF) in intensive care unit patients, overcoming major delays in diagnosis associated with current testing methods. A study published today in Science Translational Medicine reports that the NeuroSense platform can provide near real-time readings of four key biomarkers every 27 minutes, with results accurately reflecting standard clinical measurements.
In neurological intensive care units, external ventricular drainage (EVD) systems are routinely used to temporarily assist patients with drainage of excess CSF, manage postoperative complications and monitor intracranial pressure. However, the use of these devices carries a high infection risk, with rates reaching up to 20% of patients.
Delayed diagnosis of these infections can lead to severe meningitis, neural damage, cognitive impairment, permanent disability, or even death. However, current testing methods are labor-intensive and require sending samples to external laboratories for biomarker analysis and manual inspection. This limits testing to every one to two days, significantly delaying clinical decisions that can be critical for preventing severe complications.
“To address these limitations, we developed NeuroSense, a multiplexed sensing platform that integrates with standard external ventricular drainage systems to enable near real-time monitoring of key CSF biomarkers, including glucose, lactate, pH, and flow rate, that are essential for detecting infection and drain dysfunction,” write the study authors.
The NeuroSense platform employs aptamer-based biosensors to detect glucose and lactate levels in CSF, which are key markers of bacterial infections. These types of biosensors are more stable and have a longer shelf life than conventional enzymatic biosensors, ensuring the platform can consistently and accurately track these markers for the entire time EVD systems remain in place, typically between five to 10 days.
Furthermore, an impedance-based sensor measures CSF flow rate to monitor for potential catheter obstructions or incorrect EVD settings, while a polydopamine sensor keeps track of pH changes, which can indicate acidosis, hemorrhage, infection, or a disrupted blood-brain barrier.
The platform’s performance was evaluated in a small-scale study that recruited six patients with EVDs hospitalized in the intensive care unit. Every four hours, readings from the NeuroSense platform were compared with those from standard testing methods, revealing a strong correlation between the sensor platform and clinical reference measurements.
A survey of the healthcare providers and clinicians involved in the study further showed that most participants found the platform easy to use, as it integrates with standard EVD systems routinely used in the intensive care setting.
Going forward, the researchers plan on further improving the performance of the pH sensor and continue developing the platform to comply with regulatory requirements for running large-scale clinical studies and eventually making the platform available to healthcare providers.
“Beyond infection detection and EVD assessment, NeuroSense enables higher temporal-resolution tracking of CSF biomarkers and flow dynamics, supporting earlier recognition of evolving trends that may be missed with intermittent sampling,” write the researchers. “Although the current system measures glucose, lactate, pH, and flow, the platform is modular and can accommodate additional sensors in future iterations. By providing near-bedside, actionable insights into patients’ neurological health, NeuroSense has strong potential to enhance clinical decision-making and improve patient care.”
The post CSF Platform Enables Near Real-Time Monitoring of Multiple Biomarkers appeared first on Inside Precision Medicine.