Zhenghao Shi

Lung Segmentation in Chest Radiographs by Means of Gaussian Kernel-Based FCM with Spatial Constraints

A Gaussian kernel-based fuzzy clustering algorithm with spatial constraints for automatic segmentation of lung field in chest radiographs is proposed in this paper. The algorithm is realized by modifying the objective function in the conventional fuzzy c-means algorithm using Gaussian kernel-induced distance metric. The influence of the neighboring pixels on the centre pixel in chest radiograph was also taken into account to make a spatial penalty term. The methods have been tested on a publicly available database of 52 chest radiographs, in which all objects have been manually segmented by a human observer specializing in medical image analysis. Experimental results demonstrate that the proposed method is efficient and effective

An efficient active set method for optimization extreme learning machines

In this paper an efficient active set algorithm is presented for fast training of Optimization Extreme Learning Machines (OELMs). This algorithm suggests the use of an efficient identification technique of active set and the value reassignment technique for quadratic programming problem. With these strategies, this algorithm is able to drop many constraints from the active set at each iteration, and it can converge to the optimal solution with less iterations. The global convergence properties of the algorithm as well as its theoretical properties are analyzed. The effectiveness of the algorithm is demonstrated via benchmark datasets from many sources. Experiment results indicate that the quadratic programming problem which keeps the number of constraints in the active set as small as possible is computationally most efficient.

An integrated method for ancient Chinese tablet images de-noising based on assemble of multiple image smoothing filters

There are unavoidably lots of noises in tablet images due to natural or man-made decay, which have a significant affect on learning and studying of the ancient Chinese calligraphy works with Chinese tablet images. To address this problem, an integrated de-noising method, based on assemble of multiple image smoothing filters, is proposed in this paper. To avoid damaging characters and losing detail information, input Chinese tablet images are enhanced by the Guided filter and multi-scale Retinex filter firstly. Then the enhanced tablet images are converted to binary ones by the Otsu thresholding filter. Finally, most random and block noises are removed using an improved scan-length statistics filter based on connected region. The performance of the proposed method was validated on our Chinese tablet image data set, which consists of 200 Chinese tablet images with different kinds of noise. Experiments show that, the proposed method can effectively remove most image noise (including various block noise, linear noise and ant-like noise) and preserve characters better than existing methods.

A new method based on MTANNs for cutting down false-positives: An evaluation on different versions of commercial pulmonary nodule detection CAD software

One of the major problems for computer-aided pulmonary nodule detection in chest radiographs is that a high false-positive (FP) rate exists. In an effort to overcome this problem, a new method based on the MTANN (Massive Training Artificial Neural Network) is proposed in this paper. An MTANN comprises a multi-layer neural network where a linear function rather than a sigmoid function is used as its activity function in the output layer. In this work, a mixture of multiple MTANNs were employed rather than only a single MTANN. 50 MTANNs for 50 different types of FPs were prepared firstly. Then, several effective MTANNs that had higher performances were selected to construct the MTANNs mixture. Finally, the outputs of the multiple MTANNs were combined with a mixing neural network to reduce various different types of FPs. The performance of this MTANNs mixture in FPs reduction is validated on three different versions of commercial CAD software with a validation database consisting of 52 chest radiographs. Experimental results demonstrate that the proposed MTANN approach is useful in cutting down FPs in different CAD software for detecting pulmonary nodules in chest radiographs.

A deep CNN based transfer learning method for false positive reduction

A low false positive (FP) rate is of great importance for the use of a Computer Aided Detection (CAD) system to detect pulmonary nodules in thoracic Computed Tomography (CT). However, due to the variations of nodules in appear and size, it is still a very challenging task to obtain a low FP rate. In this paper, we propose a deep Convolutional Neural Network (CNN) based transfer learning method for FP reduction in pulmonary nodule detection on CT slices. We utilized one of the state-of-the-art CNN models, VGG-16 [4], as a feature extractor to obtain nodule features, and used a support vector machine (SVM) for nodule classification. Firstly we transferred all the layers from a pre-trained VGG-16 model in ImageNet to our target networks. Then, we tuned the last fully connected layers to adjust the computer-vision-task-trained CNN model to pulmonary nodule classification task. The initial CNN filter weights were then optimized using the training data, i.e., the pulmonary nodule patch images and corresponding labels through back-propagation so that they better reflected the modalities in the pulmonary nodule image dataset. Finally, features learned in the fine-tuned CNN were used to train a SVM classifier. The output of the trained SVM was used for final classification. Experimental results show that the overall sensitivity of the proposed method was 87.2% with 0.39 FPs per scan, which is higher than 85.4% with 4 FPs per scan obtained by other state of art method.

A Chinese character structure preserved denoising method for Chinese tablet calligraphy document images based on KSVD dictionary learning

As an art form, Chinese ancient calligraphy tablet works occupy an important position in the heritage of Chinese culture. However, because of natural or man-made decay, there appear lots of noises in these ancient tablet works images, which have an important effect on the quality of the tablet images. To address this problem, a character structure preserved denoising method based on KSVD dictionary learning was proposed in this paper. This new proposed method consists of two major operations: dividing-frequency denoising and ant-like noises removal in binary image. At the stage of dividing-frequency denosing, the Butterworth low pass filter was employed to filter and extract low frequency part of the images firstly. Then, KSVD dictionary learning algorithm was used for smoothing the high frequency image and extracting image edges and the extracted edge images was then fused with the denoised low frequency part of the images. At the stage of ant-like noises removal, the fused image is converted into a binary one firstly. Then, the connected region method is employed to remove isolated ant-like noise; ergodic method is used to fill holes of strokes. Finally strokes thorn was eliminated by using the median filter. Experimental results demonstrate that the proposed method can effectively remove most image noise (including various block noise, linear noise and ant-like noise) and preserve characters better than existing methods.

Automatic Building Extraction from Terrestrial Laser Scanning Data

of point clouds with different local densities, especially in the presence of random noisy points, is still a formidable challenge. In this paper, we present a complete strategy for building extraction from terrestrial laser scanning data. First, a novel segmentation method is proposed to facilitate the task of building extraction. The points are grouped based on the normals and the adjacency relationships. Second, the planar surfaces are recognized from the segmentation results based on the properties of the Gaussian image. Finally, the buildings are extracted from the urban point clouds based on a collection of characteristics of point cloud segments like shape, normal direction and topological relationship. Experimental results demonstrate that the proposed method can be used as a robust way to extract buildings from terrestrial laser scanning data. At the same time, the buildings are decomposed into several patches which lay a good foundation for building reconstruction.

Tree branching reconstruction from unilateral point clouds

Trees are ubiquitous in natural environment and realistic models of tree are also indispensable in computer graphics and virtual reality domains. However, their complexity in geometry and topology make it a great challenge for photo-realistic tree reconstruction. Since tree trunk is the preliminary structure of trees, its modeling is a critical step which plays an important role in tree modeling. Many existing methods focus on the overall resemblance of tree branches but omit the local geometry details. In this paper, we perform unilateral scanning of real-world trees and propose an approach that could reconstruct trees from incomplete point clouds. The core of our method contains four parts: local optimal segmentation of tree branch, skeletal point and lines extraction from unilateral branch, the cross-section construction of tree branch, and final tree branch surface generation. Experimental results demonstrate the effectiveness and robustness of our method which could keep realistic shape of trees.

Enhancement of Chest Radiograph Based on Wavelet Transform

The effect of anatomical noise is one of the major challenges for the early detection of pulmonary nodules in chest radiograph. A method aimed at eliminating these anatomical noises while enhancing contrast of anatomical feature is presented. The method is based on local modification of gradient magnitude values provided by the redundant dyadic wavelet transform. It includes two key steps. The first one is to threshold wavelet coefficients, which is accomplished by using a threshold strategy. The purpose of this operation is to reduce the effect of background and anatomical noise on the region of interesting in the chest radiograph. The second one is to do a normalization operation for all retained wavelet coefficients at a same scale. The purpose of this operation is to ensure that the enhanced image is not sensitive to the variance of radiograph acquirement environment. Experimental results (performed under different conditions.) indicate the efficiency and the effectiveness of the proposed method in radiography enhancement.

Reducing FPs in Nodule Detection Using Neural Networks Ensemble

In this paper, we employed neural network ensemble for FPs reduction in detecting lung nodules in chest radiographs. In our scheme, the ensemble consists of four modified forward neural networks, each one of them was trained with the back propagation algorithm to distinct a different type of non-nodules from nodules. The outputs of all the individual neural networks were combined by a modified forward mixing ANN. The performance of our scheme for false positive reduction was evaluated by use of FROC. With neural network ensemble, the false positive rate of CAD scheme1 was reduced for 44% (from 2.86 to 1.6 positives per image), at an overall sensitivity of 60%. We also compared our scheme with other researches. The result demonstrates the superiority of it over other ones. We believe that the proposed method is useful in false positives reduction in the diagnosis of lung nodules in chest radiograph.