A5082216926- Artificial Intelligence in Healthcare and Education
- COVID-19 diagnosis using AI
- Medical Imaging and Analysis
- Machine Learning in Healthcare
- Radiomics and Machine Learning in Medical Imaging
- Autopsy Techniques and Outcomes
- Ultrasound in Clinical Applications
- Radiation Dose and Imaging
- Bone fractures and treatments
- Acute Kidney Injury Research
- Hip and Femur Fractures
- Vascular Procedures and Complications
- Lower Extremity Biomechanics and Pathologies
- Liver Disease Diagnosis and Treatment
- Radiology practices and education
- Shoulder Injury and Treatment
- Sports injuries and prevention
- AI in cancer detection
- Inflammatory Bowel Disease
- Microscopic Colitis
- Topic Modeling
- Pelvic and Acetabular Injuries
- Hepatitis Viruses Studies and Epidemiology
- Sepsis Diagnosis and Treatment
- Tendon Structure and Treatment
Columbia University Irving Medical Center
2022-2025
New York University
2024-2025
Columbia University
2019-2024
NYU Langone Health
2024
St. Jude Children's Research Hospital
2024
University of Maryland, Baltimore
2024
Drexel University
2024
Wake Forest University
2024
Harvard University
2024
Boston Children's Hospital
2024
Early results in using convolutional neural networks (CNNs) on x-rays to diagnose disease have been promising, but it has not yet shown that models trained from one hospital or group of hospitals will work equally well at different hospitals. Before these tools are used for computer-aided diagnosis real-world clinical settings, we must verify their ability generalize across a variety systems. A cross-sectional design was train and evaluate pneumonia screening CNNs 158,323 chest NIH...
Hip fractures are a leading cause of death and disability among older adults. also the most commonly missed diagnosis on pelvic radiographs, delayed leads to higher cost worse outcomes. Computer-aided (CAD) algorithms have shown promise for helping radiologists detect fractures, but image features underpinning their predictions notoriously difficult understand. In this study, we trained deep-learning models 17,587 radiographs classify fracture, 5 patient traits, 14 hospital process...
This study trained long short-term memory (LSTM) recurrent neural networks (RNNs) incorporating an attention mechanism to predict daily sepsis, myocardial infarction (MI), and vancomycin antibiotic administration over two week patient ICU courses in the MIMIC-III dataset. These models achieved next-day predictive AUC of 0.876 for 0.823 MI, 0.833 administration. Attention maps built from these highlighted those times when input variables most influenced predictions could provide a degree...
Purpose To compare different methods for generating features from radiology reports and to develop a method automatically identify findings in these reports. Materials Methods In this study, 96 303 head computed tomography (CT) were obtained. The linguistic complexity of was compared with that alternative corpora. Head CT preprocessed, machine-analyzable constructed by using bag-of-words (BOW), word embedding, Latent Dirichlet allocation-based approaches. Ultimately, 1004 manually labeled...
Abstract Background Obesity is associated with progression of inflammatory bowel disease (IBD). Visceral adiposity may be a more meaningful measure obesity compared traditional measures such as body mass index (BMI). This study visceral vs BMI predictors time to IBD flare among patients Crohn’s and ulcerative colitis. Methods was retrospective cohort study. were included if they had colonoscopy computed tomography (CT) scan within 30-day window an flare. They followed for 6 months or until...
Background Rapid real-time magnetic resonance (MR) sequences enable dynamic articular kinematic assessment. The abduction-external rotation (ABER) position has long been used to characterize glenohumeral pathology. Purpose To evaluate a gradient recall echo (GRE) sequence for ABER-positioned joint assessment correlating with subjective instability and clinical apprehension testing. Material Methods Symptomatic patients were scanned using routine MR arthrogram protocol supplemented by an...
Homeless patients experience poor health outcomes and consume a disproportionate amount of care resources compared with domiciled patients. There is increasing interest in the federal government providing coordination for homeless patients, which will require systematic way identifying these individuals.We analyzed address data from Healthix, New York City-based information exchange, to identify patterns that could indicate homelessness.Patients were categorized as likely be if they...
Natural language processing tools for chest radiograph report annotation show high overall accuracy but exhibit age-related bias, with poorer performance in older patients.
Purpose To determine if weakly supervised learning with surrogate metrics and active transfer can hasten clinical deployment of deep models. Materials Methods By leveraging Liver Tumor Segmentation (LiTS) challenge 2017 public data (n = 131 studies), natural language processing reports, an method, a model was trained to segment livers on 239 retrospectively collected portal venous phase abdominal CT studies obtained between January 1, 2014, December 31, 2016. Absolute volume differences...
Summary Health information exchange (HIE) facilitates the of patient across different healthcare organizations. To match records sites, HIEs usually rely on a master index (MPI), database responsible for determining which medical at facilities belong to same patient. A single patient’s may be improperly split multiple profiles in MPI. We investigated how often two individuals shared first name, last and date birth Social Security Death Master File (SSDMF), US government containing over 85...
Patella alta (PA) and patella baja (PB) affect 1-2% of the world population, but are often underreported, leading to potential complications like osteoarthritis. The Insall-Salvati ratio (ISR) is commonly used diagnose patellar height abnormalities. Artificial intelligence (AI) keypoint models show promising accuracy in measuring detecting these abnormalities.An AI model developed validated study on a random population sample lateral knee radiographs. A was trained internally with 689...
Background: Errors in grammar, spelling, and usage radiology reports are common. To automatically detect inappropriate insertions, deletions, substitutions of words reports, we proposed using a neural sequence-to-sequence (seq2seq) model. Methods: Head CT chest radiograph from Mount Sinai Hospital (MSH) (n=61,722 818,978, respectively), Queens (MSQ) (n=30,145 194,309, respectively) MIMIC-III (n=32,259 54,685) were converted into sentences. Insertions, substitutions, deletions randomly...
Radiologists have used algorithms for Computer-Aided Diagnosis (CAD) decades. These use machine learning with engineered features, and there been mixed findings on whether they improve radiologists' interpretations. Deep offers superior performance but requires more training data has not evaluated in joint algorithm-radiologist decision systems.We developed the Note Interface (CANDI) collaboratively annotating radiographs evaluating how alter human interpretation. The annotation app collects...