A5003735451- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- COVID-19 diagnosis using AI
- Colorectal Cancer Screening and Detection
- Machine Learning in Healthcare
- Topic Modeling
- Image Retrieval and Classification Techniques
- Artificial Intelligence in Healthcare and Education
- Biomedical Text Mining and Ontologies
- Statistical Methods and Inference
- Bayesian Methods and Mixture Models
- Advanced Image and Video Retrieval Techniques
- Machine Learning and Data Classification
- Advanced Neural Network Applications
- Sepsis Diagnosis and Treatment
- Domain Adaptation and Few-Shot Learning
- Digital Radiography and Breast Imaging
- Random Matrices and Applications
- Target Tracking and Data Fusion in Sensor Networks
- Pancreatitis Pathology and Treatment
- Knee injuries and reconstruction techniques
- Lung Cancer Diagnosis and Treatment
- Probability and Risk Models
- Effects and risks of endocrine disrupting chemicals
- Total Knee Arthroplasty Outcomes
NYU Langone Health
2024
New York University
2017-2023
Cornell University
2023
Courant Institute of Mathematical Sciences
2020
Jiangsu University
2019
University of Nevada, Reno
2018
We present a deep convolutional neural network for breast cancer screening exam classification, trained, and evaluated on over 200000 exams (over 1000000 images). Our achieves an AUC of 0.895 in predicting the presence breast, when tested population. attribute high accuracy to few technical advances. 1) network's novel two-stage architecture training procedure, which allows us use high-capacity patch-level learn from pixel-level labels alongside learning macroscopic breast-level labels. 2) A...
<h3>Importance</h3> Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography accuracy by reducing missed cancers and false positives. <h3>Objective</h3> To evaluate whether AI can overcome interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. <h3>Design, Setting, Participants</h3> In this diagnostic study conducted between September 2016 November 2017, an...
Advances in deep learning for natural images have prompted a surge of interest applying similar techniques to medical images. The majority the initial attempts focused on replacing input convolutional neural network with image, which does not take into consideration fundamental differences between these two types Specifically, fine details are necessary detection images, unlike where coarse structures matter most. This difference makes it inadequate use existing architectures developed...
Medical images differ from natural in significantly higher resolutions and smaller regions of interest. Because these differences, neural network architectures that work well for might not be applicable to medical image analysis. In this work, we propose a novel model address unique properties images. This first uses low-capacity, yet memory-efficient, on the whole identify most informative regions. It then applies another higher-capacity collect details chosen Finally, it employs fusion...
Background The methods for assessing knee osteoarthritis (OA) do not provide enough comprehensive information to make robust and accurate outcome predictions. Purpose To develop a deep learning (DL) prediction model risk of OA progression by using radiographs in patients who underwent total replacement (TKR) matched control did undergo TKR. Materials Methods In this retrospective analysis that used data from the Initiative, DL on was developed predict both likelihood patient undergoing TKR...
Abstract Though consistently shown to detect mammographically occult cancers, breast ultrasound has been noted have high false-positive rates. In this work, we present an AI system that achieves radiologist-level accuracy in identifying cancer images. Developed on 288,767 exams, consisting of 5,442,907 B-mode and Color Doppler images, the area under receiver operating characteristic curve (AUROC) 0.976 a test set 44,755 exams. retrospective reader study, higher AUROC than average ten...
Abstract During the coronavirus disease 2019 (COVID-19) pandemic, rapid and accurate triage of patients at emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction deterioration risk using deep neural network that learns from chest X-ray images gradient boosting model routine clinical variables. Our AI prognosis system, trained data 3661 patients, achieves an area under receiver operating characteristic curve (AUC) 0.786 (95% CI:...
Deep learning in the presence of noisy annotations has been studied extensively classification, but much less segmentation tasks. In this work, we study dynamics deep networks trained on inaccurately annotated data. We observe a phenomenon that previously reported context classification: tend to first fit clean pixel-level labels during an "early-learning" phase, before eventually memorizing false annotations. However, contrast memorization does not arise simultaneously for all semantic...
Learning representations for individual instances when only bag-level labels are available is a fundamental challenge in multiple instance learning (MIL). Recent works have shown promising results using contrastive self-supervised (CSSL), which learns to push apart corresponding two different randomly-selected instances. Unfortunately, real-world applications such as medical image classification, there often class imbalance, so mostly belong the same majority class, precludes CSSL from...
Breast density classification is an essential part of breast cancer screening. Although a lot prior work considered this problem as task for learning algorithms, to our knowledge, all them used small and not clinically realistic data both training evaluation their models. In work, we explored the limits with set coming from over 200,000 screening exams. We train evaluate strong convolutional neural network classifier. reader study, found that model can perform comparably human expert.
In the last few years, deep learning classifiers have shown promising results in image-based medical diagnosis. However, interpreting outputs of these models remains a challenge. cancer diagnosis, interpretability can be achieved by localizing region input image responsible for output, i.e. location lesion. Alternatively, segmentation or detection trained with pixel-wise annotations indicating locations malignant lesions. Unfortunately, acquiring such labels is labor-intensive and requires...
We present a deep convolutional neural network for breast cancer screening exam classification, trained and evaluated on over 200,000 exams (over 1,000,000 images). Our achieves an AUC of 0.895 in predicting whether there is the breast, when tested population. attribute high accuracy our model to two-stage training procedure, which allows us use very high-capacity patch-level learn from pixel-level labels alongside learning macroscopic breast-level labels. To validate model, we conducted...
Breast ultrasound plays a pivotal role in detecting and diagnosing breast abnormalities. Radiology reports summarize key findings from these examinations, highlighting lesion characteristics malignancy assessments. However, extracting this critical information is challenging due to the unstructured nature of radiology reports, which often exhibit varied linguistic styles inconsistent formatting. While proprietary LLMs like GPT-4 effectively retrieve information, they are costly raise privacy...
Abstract Bovine mammary epithelial cells (MAC‐Ts) are a common cell line for the study of inflammation; these used to mechanistically elucidate molecular underpinnings that contribute bovine mastitis. mastitis is most prevalent form disease in dairy cattle culminates annual losses two billion dollars US industry. Thus, there an urgent need improved therapeutic strategies. Histone deacetylase (HDAC) inhibitors efficacious rodent models inflammation, yet their role remain unclear. HDACs have...
Abstract Pathology reports are considered the gold standard in medical research due to their comprehensive and accurate diagnostic information. Natural language processing (NLP) techniques have been developed automate information extraction from pathology reports. However, existing studies suffer two significant limitations. First, they typically frame tasks as report classification, which restricts granularity of extracted Second, often fail generalize unseen variations language, negation,...
Abstract Ultrasound is an important imaging modality for the detection and characterization of breast cancer. Though consistently shown to detect mammographically occult cancers, especially in women with dense breasts, ultrasound has been noted have high false-positive rates. In this work, we present artificial intelligence (AI) system that achieves radiologist-level accuracy identifying cancer images. To develop validate system, curated a dataset consisting 288,767 exams from 143,203...
Radiologists typically compare a patient's most recent breast cancer screening exam to their previous ones in making informed diagnoses. To reflect this practice, we propose new neural network models that pairs of mammograms from the same patient. We train and evaluate our proposed on over 665,000 images (over 166,000 exams). Our best model achieves an AUC 0.866 predicting malignancy patients who underwent screening, reducing error rate corresponding baseline.
Transformer-based detectors have shown success in computer vision tasks with natural images. These models, exemplified by the Deformable DETR, are optimized through complex engineering strategies tailored to typical characteristics of scenes. However, medical imaging data presents unique challenges such as extremely large image sizes, fewer and smaller regions interest, object classes which can be differentiated only subtle differences. This study evaluates applicability these...