Zhentai Lu

Cascade of multi-scale convolutional neural networks for bone suppression of chest radiographs in gradient domain

Suppression of bony structures in chest radiographs (CXRs) is potentially useful for radiologists and computer-aided diagnostic schemes. In this paper, we present an effective deep learning method for bone suppression in single conventional CXR using deep convolutional neural networks (ConvNets) as basic prediction units. The deep ConvNets were adapted to learn the mapping between the gradients of the CXRs and the corresponding bone images. We propose a cascade architecture of ConvNets (called CamsNet) to refine progressively the predicted bone gradients in which the ConvNets work at successively increased resolutions. The predicted bone gradients at different scales from the CamsNet are fused in a maximum-a-posteriori framework to produce the final estimation of a bone image. This estimation of a bone image is subtracted from the original CXR to produce a soft-tissue image in which the bone components are eliminated. Our method was evaluated on a dataset that consisted of 504 cases of real two-exposure dual-energy subtraction chest radiographs (404 cases for training and 100 cases for test). The results demonstrate that our method can produce high-quality and high-resolution bone and soft-tissue images. The average relative mean absolute error of the produced bone images and peak signal-to-noise ratio of the produced soft-tissue images were 3.83% and 38.7dB, respectively. The average bone suppression ratio of our method was 83.8% for the CXRs with pixel sizes of nearly 0.194mm. Furthermore, we apply the trained CamsNet model on the CXRs acquired by various types of X-ray machines, including scanned films, and our method can also produce visually appealing bone and soft-tissue images.

A robust medical image segmentation method using KL distance and local neighborhood information

In this paper, we propose an improved Chan-Vese (CV) model that uses Kullback-Leibler (KL) distances and local neighborhood information (LNI). Due to the effects of heterogeneity and complex constructions, the performance of level set segmentation is subject to confounding by the presence of nearby structures of similar intensity, preventing it from discerning the exact boundary of the object. Moreover, the CV model cannot usually obtain accurate results in medical image segmentation in cases of optimal configuration of controlling parameters, which requires substantial manual intervention. To overcome the above deficiency, we improve the segmentation accuracy by the usage of KL distance and LNI, thereby introducing the image local characteristics. Performance evaluation of the present method was achieved through experiments on the synthetic images and a series of real medical images. The extensive experimental results showed the superior performance of the proposed method over the state-of-the-art methods, in terms of both robustness and efficiency.

Retrieval of brain tumors with region-specific bag-of-visual-words representations in contrast-enhanced MRI images.

A content-based image retrieval (CBIR) system is proposed for the retrieval of T1-weighted contrast-enhanced MRI (CE-MRI) images of brain tumors. In this CBIR system, spatial information in the bag-of-visual-words model and domain knowledge on the brain tumor images are considered for the representation of brain tumor images. A similarity metric is learned through a distance metric learning algorithm to reduce the gap between the visual features and the semantic concepts in an image. The learned similarity metric is then used to measure the similarity between two images and then retrieve the most similar images in the dataset when a query image is submitted to the CBIR system. The retrieval performance of the proposed method is evaluated on a brain CE-MRI dataset with three types of brain tumors (i.e., meningioma, glioma, and pituitary tumor). The experimental results demonstrate that the mean average precision values of the proposed method range from 90.4% to 91.5% for different views (transverse, coronal, and sagittal) with an average value of 91.0%.

Automated interventricular septum segmentation for black-blood myocardial T2* measurement in thalassemia.

To develop and validate an automated segmentation method that extracts the interventricular septum (IS) from myocardial black‐blood images for the T2* measurement in thalassemia patients.

Intensity based image registration by minimizing exponential function weighted residual complexity

In this paper, we propose a novel intensity-based similarity measure for medical image registration. Traditional intensity-based methods are sensitive to intensity distortions, contrast agent and noise. Although residual complexity can solve this problem in certain situations, relative modification of the parameter can generate dramatically different results. By introducing a specifically designed exponential weighting function to the residual term in residual complexity, the proposed similarity measure performed well due to automatically weighting the residual image between the reference image and the warped floating image. We utilized local variance of the reference image to model the exponential weighting function. The proposed technique was applied to brain magnetic resonance images, dynamic contrast enhanced magnetic resonance images (DCE-MRI) of breasts and contrast enhanced 3D CT liver images. The experimental results clearly indicated that the proposed approach has achieved more accurate and robust performance than mutual information, residual complexity and Jensen-Tsallis.

Liver CT image retrieval based on non-tensor product wavelet

This paper presents a content-based medical image retrieval (CBIR) method that used in medical CT images of liver lesions with a computer-assisted diagnosis. According to medical CT images characteristics of blurred boundaries and the unconspicuous region, the liver region of interest is extracted by using semi-automatic method. We extract local co-occurrence matrix texture features and intensity features, and use improved non-tensor product wavelet filter to extract the image global features. Experimental results show that this method can improve the detection rate of lesions. It obtains good results in hepatic hemangioma and HCC which are difficult differential diagnosis both of rich blood supply to tumors.

A Fast 3-D Medical Image Registration Algorithm Based on Equivalent Meridian Plane

For the rigid registration of multi-modality medical images, mutual information (MI) technique is unsuitable to clinical diagnose because of high computational cost and low robustness. In this paper, a new concept of equivalent meridian plane (EMP) is proposed, and the EMP and other two normal feature planes are determined using principal component analysis (PCA); the rough registrations of those 2D planes are to be realized at six freedom degree; finally, the refine registrations can be completed using MI in a small neighboring region. This method is called as EMP based MI registration technique. The accuracy and robustness of EMP-MI approach can be verified by applying it to the simulated and real brain image data (CT, MR, PET, and SPECT). The experimental results indicate that the proposed algorithm reduces computational time distinctly and is a global optimal strategy.

Adaptive Segmentation of Vertebral Bodies from Sagittal MR Images Based on Local Spatial Information and Gaussian Weighted Chi-Square Distance

We present a novel method for the automatic segmentation of the vertebral bodies from 2D sagittal magnetic resonance (MR) images of the spine. First, a new affinity matrix is constructed by incorporating neighboring information, which local intensity is considered to depict the image and overcome the noise effectively. Second, the Gaussian kernel function is to weight chi-square distance based on the neighboring information, which the vital spatial structure of the image is introduced to improve the accuracy of the segmentation task. Third, an adaptive local scaling parameter is utilized to facilitate the image segmentation and avoid the optimal configuration of controlling parameter manually. The encouraging results on the spinal MR images demonstrate the advantage of the proposed method over other methods in terms of both efficiency and robustness.

Unsupervised Segmentation of Medical Image Based on FCM and Mutual Information

In the scope of medical image processing, segmentation is important and difficult. This paper presents a novel algorithm for segmentation of medical image. Our algorithm is formulated by combining the fuzzy c-means clustering (FCM) algorithm with the mutual information (MI) technique. The initial threshold can be chosen using FCM algorithm, and in the iteration process, an optimal threshold will be determined by maximizing the MI between the original volume and the thresholded volume. We evaluate the effectiveness of the proposed approach by applying it to the medical images, including magnetic resonance imaging (MRI), microphotographic image. The experimental results indicate that the proposed method has not only visually better or comparable segmentation effect but also, more favorably, removal ability for noise.

Automatic Segmentation of Myocardium from Black-Blood MR Images Using Entropy and Local Neighborhood Information

By using entropy and local neighborhood information, we present in this study a robust adaptive Gaussian regularizing Chan–Vese (CV) model to segment the myocardium from magnetic resonance images with intensity inhomogeneity. By utilizing the circular Hough transformation (CHT) our model is able to detect epicardial and endocardial contours of the left ventricle (LV) as circles automatically, and the circles are used as the initialization. In the cost functional of our model, the interior and exterior energies are weighted by the entropy to improve the robustness of the evolving curve. Local neighborhood information is used to evolve the level set function to reduce the impact of the heterogeneity inside the regions and to improve the segmentation accuracy. An adaptive window is utilized to reduce the sensitivity to initialization. The Gaussian kernel is used to regularize the level set function, which can not only ensure the smoothness and stability of the level set function, but also eliminate the traditional Euclidean length term and re-initialization. Extensive validation of the proposed method on patient data demonstrates its superior performance over other state-of-the-art methods.