Jon Wang

D

Deep Vein Thrombosis Screening with Three-Dimensional Deep Learning on Lower Extremity Computed Tomography Studies

Mar 2019 - Jun 2019 (4 months)

Jonathan X Wang*, Brianna Kozemzak*, Anoop Manjunath*, Trevor Tsue*, Andre Souffrant, and Lawrence Hoffman (*equal contributors) Recent advances in deep neural networks (DNNs) allow us to leverage spatial dependencies between slices in imaging studies to identify false positives and ultimately deploy DNN systems that lighten physicians’ workloads while not exacerbating alarm fatigue. To train our DNN, we have acquired 119 lower-body CT imaging studies labeled by radiologists for DVT at the pixel level. Using these studies, we have developed a DNN-based CAD system that will (1) segment targeted deep veins in a CT slice, (2) classify whether a DVT is present within multiple slices given segmentations of deep veins, and (3) evaluate different

D

DeepDoc: Natural Language Processing with Deep Neural Networks for the American Board of Internal Medicine Certification

Jan 2019 - Current (7 years 6 months)

We train a model to answer review questions for the American Board of Internal Medicine Certification Exam. We adapt approaches traditionally used for question answer tasks to our multiple choice exam, as well as experiment with the following enhancements: PubMed Embeddings, BiDAF, DrQA, SAR, GA, and RACE. Ultimately we find that GA models perform best (Accuracy: 0.38, AUROC: 0.64). Our work is an initial study towards the development of a intelligent medical QA system, demonstrating the capability of modern day machine learning to answer questions clinicians typically take many years to study for.

D

DeepSign: Efficient Siamese Convolutional Neural Networks for Signature Verification

Jan 2019 - Current (7 years 6 months)

We develop a deep learning algorithm that performs handwritten signature verification. We created our own SqueezeNet-inspired efficient siamese convolutional neural network architecture, DeepSign, that uses 65% fewer parameters than Google’s MobileNetv2 and 97% fewer parameters than the current state of the art, SigNet, while acheiving similar if not better performance. We test our models on both the CEDAR and BHSig260 datasets and demonstrate that our model outperforms both models in all evaluation metrics (accuracy: 0.85, precision: 0.76, recall: 0.84, AUROC: 0.93). This lightweight model is readily applicable to mobile devices for both online or offline signature verification. DeepSign is integrated into a React web app so that viewers c

T

The Role of Macrophages in Tumor Cell Recurrence Following Radiation Therapy

Sep 2015 - Jun 2019 (3 years 10 months)

Radiotherapy is an effective treatment modality for more than fifty percent of all cancer patients. However, pre-clinical evidence suggests radiotherapy increases the risk for tumor recurrence. This process may be mediated through the recruitment of circulating tumor cells (CTCs) by radiation-induced expression of the cytokine granulocyte macrophage colony stimulating factor (GM-CSF). The pathway by which GM-CSF recruits these CTCs continues to be an area of study. GM-CSF is known to promote the proliferation and recruitment of monocytes to a given location and induce their differentiation into macrophages. Macrophages have a variety of interactions with cancer and may promote tumor development or facilitate tumor cell death depending on th

M

Mobineo

Jun 2015 - Sep 2015 (4 months)

Working to digitize land in developing countries.

M

Machine Learning for Automated Classification of Patient Cases

Jonathan X Wang, Cole Deisseroth, James Bai, Jonathan H Chen (equal contributors) This is an initial study toward the development of an intelligent patient-allocation system to save medical personnel valuable time, and help patients find the care they need more efficiently by automatically categorizing cases into specific departments. We develop an algorithm which predicts the categories of patient cases from the American Board of Internal Medicine Examinations—a certification that all physicians must go through to practice general medicine. Our ontology breaks questions into their components (Case, AnswerChoice, Explanation). We then run an automatic concept extractor (ClinPhen) on the passage (description of the case) to compile a list of

A

Automated Electronic Calculator for Management of DKA/HHS

Surveyed 73 staff members to develop DKA software estimated to save $78,000 a year and reduce readmissions by 45%. Currently implemented in Stanford Hospital and integrated into the hospital’s health record system along with educational video. Awarded best poster at Resident & Fellow Quality Improvement & Patient Safety Symposium

About

Detail

Timeline

Résumé