Juan Shan

Computer-Aided Diagnosis for Breast Ultrasound Using Computerized BI-RADS Features and Machine Learning Methods

This work identifies effective computable features from the Breast Imaging Reporting and Data System (BI-RADS), to develop a computer-aided diagnosis (CAD) system for breast ultrasound. Computerized features corresponding to ultrasound BI-RADs categories were designed and tested using a database of 283 pathology-proven benign and malignant lesions. Features were selected based on classification performance using axa0bottom-up approach for different machine learning methods, including decision tree, artificial neural network, random forest and support vector machine. Using 10-fold cross-validation on the database of 283 cases, the highest area under the receiver operating characteristic (ROC) curve (AUC) was 0.84 from a support vector machine with 77.7% overall accuracy; the highest overall accuracy, 78.5%, was from a random forest with the AUC 0.83. Lesion margin and orientation were optimum features common to all of the different machine learning methods. These features can be used in CAD systems to help distinguish benign from worrisome lesions.

A Deep Learning Method for Microaneurysm Detection in Fundus Images

Diabetic Retinopathy (DR) is the leading cause of blindness in the working-age population. Microaneurysms (MAs), due to leakage from retina blood vessels, are the early signs of DR. However, automated MA detection is complicated because of the small size of MA lesions and the low contrast between the lesion and its retinal background. Recently deep learning (DL) strategies have been used for automatic feature extraction and classification problems, especially for image analysis. In this paper, a Stacked Sparse Autoencoder (SSAE), an instance of a DL strategy, is presented for MA detection in fundus images. Small image patches are generated from the original fundus images. The SSAE learns high-level features from pixel intensities alone in order to identify distinguishing features of MA. The high-level features learned by SSAE are fed into a classifier to categorize each image patch as MA or non-MA. The public benchmark DIARETDB is utilized to provide the training/testing data and ground truth. Among the 89 images, totally 2182 image patches with MA lesions, serve as positive data, and another 6230 image patches without MA lesions are generated by a randomly sliding window operation, to serve as negative data. Without any blood vessel removal or complicated preprocessing operations, SSAE learned directly from the raw image patches, and automatically extracted the distinguishing features to classify the patches using Softmax Classifier. By employing the fine-tuning operation, an improved F-measure 91.3% and an average area under the ROC curve (AUC) 96.2% were achieved using 10-fold cross-validation.

CyberCare: A Novel Electronic Health Record Management System

Computer Science is a field that runs on innovation and departs from traditional ways of completing certain tasks. Data processing has been undergoing major changes during the past few decades. While written documentation was once the only way to organize and store patient information, electronic databases have been ushered into the medical field as a tool for a better type of healthcare. This paper discusses the implementation of an EHR prototype that maintains the efficiency of existing systems, but advances away from their outdated features. The novel components of the implemented system are the voice navigation feature, medical image processing and editing ability, and interactive calendar. The newly developed health management system aims to provide an efficient and convenient tool for physicians to manage patient information and medical records.

Multiplayer tracking system for short track speed skating

In this study, the authors propose a novel and effective short track speed skating tracking system. Aimed at several challenging tracking problems of short track skating: long-time occlusion and complex group situations, size variations, similar or identical uniforms, fast motion speed, quick orientation changes etc.; a novel fuzzy model is proposed to initialise contours of skaters even they wear the same kind of uniform. Then, a novel approach for multiple object tracking is developed to track skaters reliably. The experimental results demonstrate that the proposed system can solve the above challenging problems very effectively. The information provided by the proposed system, including trajectories, velocity analysis and two-dimensional reconstruction animations, is valuable to broadcasters, athletes and coaches alike. The proposed tracking algorithm can obtain better results than that by the other eight state-of-the-art trackers in short track speed skating tracking.

A novel and robust automatic seed point selection method for breast ultrasound images

Accurate segmentation of breast lesions is among the several challenges in the development of a fully automatic Computer-Aided Diagnosis system for solid breast mass classification. Many high level segmentation methods rely heavily on proper initialization and the seed point selection is usually the necessary first step. In this paper, a fully automatic and robust seed point selection method is proposed. The method involves a number of processing steps in both space and frequency domain and endeavors to incorporate the breast anatomical knowledge. Using a database of 498 images, we compared the proposed method with two other state-of-the-art methods; the proposed method outperforms both methods significantly with a success rate of 62.85% vs. 44.97% and 13.05% on seed point select.

A new scheme to evaluate the accuracy of knowledge representation in automated breast cancer diagnosis

In the field of breast cancer diagnosis, computer-aided diagnosis (CAD) systems can provide doctors important second opinions, relying on the advanced computation ability and artificial intelligence of computing systems. The collaboration between doctors and CAD systems can help reducing the false diagnosis rate. Knowledge representation is an important chain for any artificial intelligence system, including automated breast cancer diagnosis. The breast cancer experts knowledge of distinguishing benign and malignant lesions is well described by Breast Imaging Reporting and Data System (BIRADS). Many digital formulas have been proposed to quantify BIRADS features. However, there is no direct evaluation scheme for these digital features. A common way that people evaluate digital features is using them as the input for classifiers, such as machine learning methods, and then evaluating the performance of classifiers, which indirectly serves as the evaluation of digital features. The performance of a classifier is affected by the digital features, but also affected by other factors. It is inaccurate to use only the performance of classifiers as the metric to evaluate digital features. The vision of this work is to separate the evaluation of digital features from the evaluation of classifiers, with the purpose of providing an accurate feature measurement procedure and improving the quality of knowledge representation. An independent feature evaluation scheme without using any automatic classifier is proposed. Such a scheme can directly evaluate how precisely experts knowledge is represented in computerized systems. Several commonly used digital features and newly proposed digital features in this work are evaluated using this scheme on a breast ultrasound image database. Pathological results and radiologists opinions serve as the ground truth for evaluation purpose.