Bin Kong

◦ Building a next-generation Recommender system for comment ranking, serving billions requests daily ◦ Implemented and deployed different ML model, significantly boosting daily active users (DAP) for comments creates ◦ Developed and deployed different integrity models to downrank/filter bad comments ◦ Proposed and implemented different comment features to provide better personalized ranking

Non-invasive Coronary Artery Functional Assessment from CTA Images ◦ Designed and implemented Graph Neural Networks (GNNs) to predict Fractional Flow Reserve (FFR) ◦ Proposed and implemented FFR drop-based approach to enforce monotonicity of FFR prediction ◦ Proposed and developed end-to-end sparse convolution based FFR quantification framework to predict FFR directly from CTA images ◦ Designed and Implemented lesion-based FFR quantification model to enforce lesion impact in FFR prediction ◦ Implemented prefetching, batching of coronary artery trees, accelerating training and testing by almost 5 times ◦ Introduced and developed stochastic gate-based feature selection, reducing redundant features by 50% ◦ Integrated and automated preprocessing, inference, and postprocessing steps, accelerating inference by 3 times ◦ Managing CTA dataset, FFR quantification codebase (including preprocessing such as coronary artery segmentation, centerline extraction, feature extraction, FFR quantification, and postprocessing) to ensure code quality and reproducibility ◦ Designing and improving visualization tools to facilitate FFR analysis software development and case analysis ◦ Registered more than 10 U.S. patents COVID-19 Triage Framework ◦Developed a convolutional neural network (CNN) based triage system to differentiate COVID-19 CT scans ◦Implemented the gradient-weighted class activation mapping (Grad-CAM) to visualize key image regions◦ Validated on 4352 chest CT scans, with sensitivity 90% and specificity 96% for COVID-19 classification◦Published inRadiologywith over 1,300 citations

Actively involved in many projects on medical image analysis with deep learning under supervision of Prof. Shaoting Zhang. Developed a novel CNN and 2D LSTM based framework to leverage the spatially structured information to detect cancer metastasis in pathology images (Python, Caffe, C++) Proposed a CNN and LSTM based framework for End-Diastole (ED) and End-Systole (ES) frame detection (Python, C++, Caffe) Proposed and implemented a transfer learning framework for unsupervised domain adaptation with generative adversarial network (GAN) (Python, Pytorch)

Designed and Implemented an efficient invasive cancer detection framework with transfer learning and compressed network (Pytorch, C++) Developed a Mask R-CNN network to segment the hands and used it for hand bone age assessment (PyTorch, Python)

Developed a color correction system for high resolution images (more than 100,000x100,000) cultural heritage images Developed and implemented a SIFT feature based registration and stitching system for high resolution images

Implemented and Delivered several medical image analysis softwares Proposed and implemented a novel barycenter based registration system for mamography image registration (C++, OpenCV) Developed Multi-Scale Image Contrast Amplification (MUSICA) system for X-ray image enhancement (C++, OpenCV) Implemented a mutual information based X-ray image stitching system (C++, OpenCV)