Advances in Medicine

Artificial Intelligence

Artificial intelligence is rapidly transforming health care by enhancing clinical decision-making, accelerating research, and enabling more personalized, data-driven care. NewYork-Presbyterian is at the forefront of this transformation, pioneering the integration of AI into medicine to improve outcomes, from cardiovascular diagnostics to breast cancer risk assessment and beyond. Physicians and researchers from Columbia and Weill Cornell Medicine have been working across disciplines and in the lab with data scientists and engineers to develop, validate, and deploy AI-driven solutions that elevate patient care and predictive medicine. This report showcases how clinical expertise and academic strength come together to redefine what’s possible in complex care with the help of machine learning.

Download the Full Report View the E-booklet

Read more about our latest applications of artificial intelligence:

Cardiology & Heart Surgery

AI screening tool detects structural heart disease using ECGs

Novel AI tool offers less-invasive alternative for 8 calculating FFR in coronary stenosis patients

Groundbreaking AI models transform cardiovascular diagnosis

Oncology

Deep learning helps predict treatment response in muscle-invasive bladder cancer

Algorithm analyzes breast textures on mammograms for more accurate risk assessment

Neurology & Neurosurgery

Data mining and machine learning enhance neurocritical care

Orthopedics

AI-based software improves concussion diagnostics and safety

Psychiatry

AI model may help predict onset of schizophrenia after early psychosis

Women's Health

AI clinical tool flags pregnant patients at risk for postpartum depression

Patient-driven app collects critical data to further endometriosis research

AI-powered system improves fertility success

Ophthalmology

Combining ophthalmology expertise and data science to improve diagnostics and training

AI screening tool detects structural heart disease using ECGs

A deep learning model accurately identified structural heart disease on electrocardiogram (ECG) readings at a higher frequency than cardiologists, including those who used AI to help interpret data. Pierre Elias, M.D., a cardiologist at NewYork-Presbyterian and Columbia and medical director for artificial intelligence at NewYork-Presbyterian, led a team of physicians and researchers in developing and validating EchoNext, a convolutional neural network model trained on more than 1.2 million ECG-echocardiogram pairs from 230,000 patients. It accurately identified 77% of structural heart problems compared to 64% of cardiologists on non–AI-assisted reviews and 69% for AI-assisted ones. The data supports the use of AI to improve screening access and aid physicians in early detection.

Novel AI tool offers less-invasive alternative for calculating FFR in coronary stenosis patients

A large international randomized trial led by Ajay Kirtane, M.D., interventional cardiologist at NewYork-Presbyterian and Columbia, proved that a novel software-based tool calculates fractional flow reserve (FFR) as effectively as the more-invasive wire-based standard, while also reducing procedural time. The technology, known as FFRangio,uses advanced computations guided by artificial intelligence to create a 3D model of a patient’s arteries from routine angiogram images, helping physicians determine if percutaneous coronary intervention may be needed to treat stenosis. In a trial of more than 1,900 patients across five countries, FFRangio patients experienced a major adverse cardiovascular event at one year at rates similar to those in the conventional group — demonstrating noninferiority and advancing a streamlined approach that aligns with clinical guidelines.

Groundbreaking AI models transform cardiovascular diagnosis

Pierre Elias, M.D., a cardiologist at NewYork-Presbyterian and Columbia and medical director for artificial intelligence at NewYork-Presbyterian, is at the forefront of using AI to reshape cardiovascular diagnostics. He leads the Center for Cardiovascular and Radiologic Deep Learning (CRADLE), where a multidisciplinary team of data scientists, engineers, and clinicians develop tools to help physicians identify heart issues using standard electrocardiogram (ECG) and echocardiogram data. Models developed by the CRADLE lab include DELINEATE, which improves detection of valvular heart disease; ATTRaction, which has significantly increased diagnoses of cardiac amyloidosis; and EchoNext, the first model to detect all forms of structural heart disease from ECG data alone. The technologies have already led to lifesaving interventions, giving clinicians better tools to diagnose and treat structural heart disease earlier.

Watch more

Groundbreaking AI models transform cardiovascular diagnosis

Deep learning helps predict treatment response in muscle-invasive bladder cancer

A study co-led by Bishoy Faltas, M.D., a genitourinary oncologist at NewYork-Presbyterian and Weill Cornell Medicine, explored the ability of deep learning models to enhance treatment decision-making for patients with muscle-invasive bladder cancer, potentially sparing them from surgery to remove the bladder. Dr. Faltas and his team evaluated the novel Graph-based Multimodal Late Fusion model, which integrates histopathology and cell type data, and found that it accurately predicted response to neoadjuvant chemotherapy by identifying biomarkers of response. Results show that the technology could help patients avoid a surgery that has high morbidity and mortality, demonstrating the potential of AI to personalize clinical care and improve outcomes.

Algorithm analyzes breast textures on mammograms for more accurate risk assessment

A study co-led by Despina Kontos, Ph.D., vice chair of artificial intelligence and data science research in the Department of Radiology at Columbia, shows that breast cancer risk can be predicted more accurately by analyzing breast texture patterns on mammograms rather than relying solely on breast density, the current standard. Using radiomics, researchers identified six distinct breast texture phenotypes that outperform density in risk prediction and developed a computer algorithm that can detect and quantify these patterns, which may soon be added to digital mammography systems to help clinicians with their risk estimations. The findings underscore that women with the same breast density may have very different risk profiles, necessitating clinical tools that support personalized screening strategies for early detection.

Data mining and machine learning enhance neurocritical care

Among patients recovering from aneurysmal subarachnoid hemorrhage there is a significant risk for stroke, but it can be hard to predict who is likely to develop additional complications. To help address this, Soojin Park, M.D., a neurologist and medical director of critical care data science and artificial intelligence at NewYork-Presbyterian and Columbia, has applied her experience in the neurocritical intensive care unit and her background in data science to create the Continuous Monitoring Tool for Delayed Cerebral Ischemia (COSMIC) score. On the Advances in Care podcast, she discussed how the COSMIC system uses machine learning and patient data to accurately assess risk and how it creates opportunities for targeted neurocritical care.

Listen here

Data mining and machine learning enhance neurocritical care

Al-based software improves concussion diagnostics and safety

Most concussion diagnoses are based on self-reported symptoms and subjective clinical evaluations. Along with a team of collaborators, Thomas Bottiglieri, D.O., chief of the primary care sports medicine division in the Department of Orthopedics at NewYork-Presbyterian and Columbia, has developed software that uses a computer algorithm to quantify a biometric that can help physicians diagnose concussions in a more objective way. The assessment tool uses an eye-tracking headset to capture subtle head and neck movements associated with concussive injury and can detect the biomarker with 80% to 90% sensitivity. Early clinical cases highlight the tool’s potential to guide safer return-to-play decisions and prevent repeated head trauma, transforming the way concussions are diagnosed and managed in athletes.

Listen here Read more

Novel Remote Monitoring Device May Detect Heart Failure Events Earlier

AI model may help predict onset of schizophrenia after early psychosis

To address the challenge of diagnosing schizophrenia in its early stages, a team of Columbia researchers, including Steven A. Kushner, M.D., Ph.D., a professor of psychiatry, and Shalmali Joshi, Ph.D., an assistant professor of biomedical informatics, developed a deep learning model designed to identify which patients with early psychosis are at high risk of developing schizophrenia. The model was trained on Medicaid data and revealed that the most predictive factors were not symptom reports but how patients used healthcare services, including how often a person sought care, in which settings, and the types of services used. These findings demonstrate the potential for a tool that could assist clinicians in decision-making using data that can already be found in standard electronic health records.

AI clinical tool flags pregnant patients at risk for postpartum depression

A new AI-powered prediction tool is helping physicians identify which pregnant patients may be at risk for developing postpartum depression (PPD). Rochelle Joly, M.D., an OB-GYN at NewYork-Presbyterian and Weill Cornell Medicine, and Yiye Zhang, Ph.D., an associate professor of population health sciences at Weill Cornell Medicine, led a team that developed a model that analyzes 30 variables in electronic health records to identify which patients may have PPD risk before symptoms begin. By flagging these patients with a visual alert in clinician-facing medical records, the tool both expands access to mental health screening and supports clinical decision-making by helping obstetricians assess what types of care or treatment a patient may need.

Patient-driven app collects critical data to further endometriosis research

Noémie Elhadad, Ph.D., chair of the Department of Biomedical Informatics at Columbia, is at the forefront of using citizen science to better understand endometriosis. She and her colleagues launched a long-term research initiative called Citizen Endo, which uses patient-reported data to study the disease. One arm of Citizen Endo is Phendo, a digital app developed with input from women with endometriosis, who can use it to report and track their symptoms and daily experiences. Working with clinical partners including Arnold Advincula, M.D., chief of gynecologic specialty surgery at NewYork-Presbyterian and Columbia, Dr. Elhadad and colleagues are using the data to conduct research on the systemic nature of endometriosis and leveraging AI to help identify characteristics and phenotypes of the disease. Her team is also working on future projects that will employ AI to analyze electronic health data to identify patients at risk for endometriosis in the hopes of improving early detection and diagnosis.

AI-powered system improves fertility success

On the Advances in Care podcast, Zev Williams, M.D., Ph.D., chief of the Division of Reproductive Endocrinology and Fertility at NewYork-Presbyterian and Columbia, discussed a groundbreaking AI-powered tool he and his team developed to tackle infertility caused by azoospermia, the absence of detectable sperm in a semen sample. The Sperm Track and Recovery (STAR) tool integrates advanced imaging and microfluidic chip technology with AI to accurately identify and capture trace amounts of sperm for potential future fertilization. In early cases, STAR was able to find sperm within an hour that humans could not find over two days, showcasing the transformative potential of AI in reproductive medicine and its ability to improve fertility success.

Listen here

AI-powered system improves fertility success

Combining ophthalmology expertise and data science to improve diagnostics and training

The Artificial Intelligence for Vision Science (AI4VS) Lab combines the expertise of NewYork-Presbyterian ophthalmic specialists and Columbia biomedical engineers and data scientists to research AI-based solutions that enhance clinical decision-making and patient care. Lab investigators, under the direction of Kaveri Thakoor, Ph.D., are spearheading the development of AI models to support physicians across all areas of practice. Examples of these initiatives include building AI-assisted screening for thyroid eye disease with oculoplastic surgeon Lora Dagi Glass, M.D., and creating an AI training model with Royce Chen, M.D., residency director for the Department of Ophthalmology, that uses gaze-tracking data to improve how ophthalmic trainees diagnose eye diseases through imaging. These clinical-academic partnerships leverage the wealth of data and imaging that is used in ophthalmic practice by turning it into real-world ways to help patients.

Stay up to date with the Advances newsletter

Get the latest groundbreaking research and advancements in medicine delivered to your inbox.