融合Vnet和边缘特征的肺结节分割算法

目的:针对肺结节分割过程中存在边缘信息丢失、边界分割模糊问题,在Vnet基础上改进BSR-Vnet算法。方法:首先,利用边缘关键点选择算法和空洞卷积改进BFB边缘特征增强模块,集成进编码器中,用以保留更多边缘信息;其次,用空间-通道压缩激励模块sc-SE引入双注意力机制,代替Vnet的瓶颈结构,提取非局部上下文信息;最后,利用Vnet原有的残差结构,使用混合空洞卷积构造残差空洞模块RDB代替原本的解码器,用于扩大感受野,提取更多特征细节。结果:本文改进的BSR-Vnet应用在公共数据集LIDC-IDRI上,在Dice系数上达到了87.94%的结果。结论:与基础Vnet方法相比,该模型有效保留了更多的边缘结构和信息,提取了更多上下文信息,使肺结节分割结果更为精确。 【关键词】肺结节分割;Vnet网络;空洞卷积;注意力机制;边缘关键点选择算法     

Segment-based dose optimization using a genetic algorithm

Intensity modulated radiation therapy (IMRT) inverse planning is conventionally done in two steps. Firstly, the intensity maps of the treatment beams are optimized using a dose optimization algorithm. Each of them is then decomposed into a number of segments using a leaf-sequencing algorithm for delivery. An alternative approach is to pre-assign a fixed number of field apertures and optimize directly the shapes and weights of the apertures. While the latter approach has the advantage of eliminating the leaf-sequencing step, the optimization of aperture shapes is less straightforward than that of beamlet-based optimization because of the complex dependence of the dose on the field shapes, and their weights. In this work we report a genetic algorithm for segment-based optimization. Different from a gradient iterative approach or simulated annealing, the algorithm finds the optimum solution from a population of candidate plans. In this technique, each solution is encoded using three chromosomes: one for the position of the left-bank leaves of each segment, the second for the position of the right-bank and the third for the weights of the segments defined by the first two chromosomes. The convergence towards the optimum is realized by crossover and mutation operators that ensure proper exchange of information between the three chromosomes of all the solutions in the population. The algorithm is applied to a phantom and a prostate case and the results are compared with those obtained using beamlet-based optimization. The main conclusion drawn from this study is that the genetic optimization of segment shapes and weights can produce highly conformal dose distribution. In addition, our study also confirms previous findings that fewer segments are generally needed to generate plans that are comparable with the plans obtained using beamlet-based optimization. Thus the technique may have useful applications in facilitating IMRT treatment planning.     

Characterization of mammographic masses using a gradient-based segmentation algorithm and a neural classifier

Computerized methods have recently shown a great potential in providing radiologists with a second opinion about the visual diagnosis of the malignancy of mammographic masses. The computer-aided diagnosis (CAD) system we developed for the mass characterization is mainly based on a segmentation algorithm and on the neural classification of several features computed on the segmented mass. Mass-segmentation plays a key role in most computerized systems. Our technique is a gradient-based one, showing the main characteristic that no free parameters have been evaluated on the data set used in this analysis, thus it can directly be applied to data sets acquired in different conditions without any ad hoc modification. A data set of 226 masses (109 malignant and 117 benign) has been used in this study. The segmentation algorithm works with a comparable efficiency both on malignant and benign masses. Sixteen features based on shape, size and intensity of the segmented masses are extracted and analyzed by a multi-layered perceptron neural network trained with the error back-propagation algorithm. The capability of the system in discriminating malignant from benign masses has been evaluated in terms of the receiver-operating characteristic (ROC) analysis. A feature selection procedure has been carried out on the basis of the feature discriminating power and of the linear correlations interplaying among them. The comparison of the areas under the ROC curves obtained by varying the number of features to be classified has shown that 12 selected features out of the 16 computed ones are powerful enough to achieve the best classifier performances. The radiologist assigned the segmented masses to three different categories: correctly-, acceptably- and non-acceptably-segmented masses. We initially estimated the area under ROC curve only on the first category of segmented masses (the 88.5% of the data set), then extending the classification to the second subclass (reaching the 97.8% of the data set) and finally to the whole data set, obtaining A(z)=0.805+/-0.030, 0.787+/-0.024 and 0.780+/-0.023, respectively.     

Volumetric measurements of right cerebral hemisphere infarction: use of a semiautomatic MRI segmentation technique

The applications of a new segmentation software, Anatomatic, in the evaluation of volumetric measurements of brain infarctions and the new Medimag 3D software in the evaluation of 3D image representation of infarctions are described. These programs are applied to magnetic resonance imaging. The aim of this study is to evaluate the use of these software packages in making accurate volumetric measurements in 40 patients with right cerebral infarctions, in determining the correlations between the quantitated lesions and neurological/neuropsychological dysfunctions and in creating realistic 3D views of the infarctions. Using Anatomatic, reproducible infarction volumes were achieved with ease and within a reasonably fast time. Medimag helped achieve realistic 3D representations of the infarctions. When compared, the semiautomatic segmentation proved to be much faster and yielded higher infarction volumes than the manual segmentation technique. Significantly positive correlations between the infarction volumes and neurological dysfunctions and neuropsychological deficit (neglect) helped to explain the effect of volumes on the clinical status of the patients.     

Multi-species prostate implant treatment plans incorporating 192Ir and 125I using a Greedy Heuristic based 3D optimization algorithm

The goals of interstitial implant brachytherapy include delivery of the target dose in a uniform manner while sparing sensitive structures, and minimizing the number of needles and sources. We investigated the use of a multi-species source arrangement (192Ir with 125I) for treatment in interstitial prostate brachytherapy. The algorithm utilizes an "adjoint ratio," which provides a means of ranking source positions and is the criterion for the Greedy Heuristic optimization. Three cases were compared, each using 0.4 mCi 125I seeds: case I is the base case using 125I alone, case II uses 0.12 mCi 192Ir seeds mixed with 125I, and case III uses 0.25 mCi 192Ir mixed with 125I. Both multi-species cases result in lower exposure of the urethra and central prostate region. Compared with the base case, the exposure to the rectum and normal tissue increases by a significant amount for case III as compared with the increase in case II, signifying the effect of slower dose falloff rate of higher energy gammas of 192Ir in the tissue. The number of seeds and needles decreases in both multi-species cases, with case III requiring fewer seeds and needles than case II. Further, the effect of 192Ir on uniformity was investigated using the 0.12 mCi 192Ir seeds in multi-species implants. An increase in uniformity was observed with an increase in the number of 0.12 mCi 1921r seeds implanted. The effects of prostate size on the evaluation parameters for multi-species implants were investigated using 0.12 mCi 192Ir and 0.4 mCi 125I, and an acceptable treatment plan with increased uniformity was obtained.     

Semiautomatic three-dimensional segmentation of the prostate using two-dimensional ultrasound images

In this paper, we report on two methods for semiautomatic three-dimensional (3-D) prostate boundary segmentation using 2-D ultrasound images. For each method, a 3-D ultrasound prostate image was sliced into the series of contiguous 2-D images, either in a parallel manner, with a uniform slice spacing of 1 mm, or in a rotational manner, about an axis approximately through the center of the prostate, with a uniform angular spacing of 5 degrees. The segmentation process was initiated by manually placing four points on the boundary of a selected slice, from which an initial prostate boundary was determined. This initial boundary was refined using the Discrete Dynamic Contour until it fit the actual prostate boundary. The remaining slices were then segmented by iteratively propagating this result to an adjacent slice and repeating the refinement, pausing the process when necessary to manually edit the boundary. The two methods were tested with six 3-D prostate images. The results showed that the parallel and rotational methods had mean editing rates of 20% and 14%, and mean (mean absolute) volume errors of -5.4% (6.5%) and -1.7% (3.1%), respectively. Based on these results, as well as the relative difficulty in editing, we conclude that the rotational segmentation method is superior.     

Automatic segmentation algorithm for the extraction of lumen region and boundary from endoscopic images

A new segmentation algorithm for lumen region detection and boundary extraction from gastro-intestinal (GI) images is presented. The proposed algorithm consists of two steps. First, a preliminary region of interest (ROI) representing the GI lumen is segmented by an adaptive progressive thresholding (APT) technique. Then, an adaptive filter, the Iris filter, is applied to the ROI to determine the actual region. It has been observed that the combined APT-Iris filter technique can enhance and detect the unclear boundaries in the lumen region of GI images and thus produces a more accurate lumen region, compared with the existing techniques. Experiments are carried out to determine the maximum error on the extracted boundary with respect to an expert-annotated boundary technique. Investigations show that, based on the experimental results obtained from 50 endoscopic images, the maximum error is reduced by up to 72 pixels for a 256 × 256 image representation compared with other existing techniques. In addition, a new boundary extraction algorithm, based on a heuristic search on the neighbourhood pixels, is employed to obtain a connected single pixel width outer boundary using two preferential sequence windows. Experimental results are also presented to justify the effectiveness of the proposed algorithm.     

Outlining the prostate boundary using the harmonics method

In a computerised ultrasound image guidance for automated prostatectomy system, it is necessary to identify a smooth, continuous contour for the prostate (boundary) from the ultrasound image. The radial bas-relief (RBR) method, which has been reported previously, can extract a skeletonised image from an ultrasound image automatically. After this process the prostate boundary is clearly revealed. However, analysis of the image is far from complete, as there are many spurious branches that create too much ambiguity for the system to define the actual boundary. There are also sections missing from the prostate boundary. Therefore further post-processing is required to describe and define the prostate boundary. In the paper, the harmonics method is used to describe the prostate boundary. The harmonics method uses Fourier information for noise removal and encodes a smooth boundary. The results of using the harmonics method after application of the RBR method on ultrasound images are presented. Factors that affect the performance are also highlighted and discussed.     

X-ray imaging optimization using virtual phantoms and computerized observer modelling

This study develops and demonstrates a realistic x-ray imaging simulator with computerized observers to maximize lesion detectability and minimize patient exposure. A software package, ViPRIS, incorporating two computational patient phantoms, has been developed for simulating x-ray radiographic images. A tomographic phantom, VIP-Man, constructed from Visible Human anatomical colour images is used to simulate the scattered portion using the ESGnrc Monte Carlo code. The primary portion of an x-ray image is simulated using the projection ray-tracing method through the Visible Human CT data set. To produce a realistic image, the software simulates quantum noise, blurring effects, lesions, detector absorption efficiency and other imaging artefacts. The primary and scattered portions of an x-ray chest image are combined to form a final image for computerized observer studies and image quality analysis. Absorbed doses in organs and tissues of the segmented VIP-Man phantom were also obtained from the Monte Carlo simulations. Approximately 25,000 simulated images and 2,500,000 data files were analysed using computerized observers. Hotelling and Laguerre-Gauss Hotelling observers are used to perform various lesion detection tasks. Several model observer tasks were used including SKE/BKE, MAFC and SKEV. The energy levels and fluence at the minimum dose required to detect a small lesion were determined with respect to lesion size, location and system parameters.     

Accuracy and variability assessment of a semiautomatic technique for segmentation of the carotid arteries from three-dimensional ultrasound images

In this paper, we report on a semiautomatic method for segmentation of three-dimensional (3D) carotid vascular ultrasound (US) images. Our method is based on a dynamic balloon model represented by a triangulated mesh. The mesh is manually placed within the interior of the carotid vessels, then is driven outward until it reaches the vessel wall by applying an inflation force to the mesh. Once the mesh is in close proximity to the vessel wall, it is further deformed using an image-based force, in order to better localize the boundary. Since the method requires manual initialization, there is inherent variability in the position and shape of the final segmented boundary. Using a 3D US image of a patient's carotids, we have examined the local variability in boundary position as the initialization position is varied throughout the interior of the carotid vessels in the 3D image. We have compared the semiautomatic segmentation method to a fully manual segmentation method, and found that the semiautomatic approach is less variable than the intraobserver variability for manual segmentation. We have furthermore examined the accuracy of the semiautomatic method by comparing the average surface to an "ideal" surface, determined by the average manually segmented surface. We have found, in general, good agreement between the semiautomatic and manual segmentation methods. For the 3D US image in question, the mean separation between the average segmented surface and the gold standard was found to be 0.35 mm. The two surfaces were determined to agree with each other, within uncertainty, at 65% of the mesh points comprising the two surfaces.     

An acoustical respiratory phase segmentation algorithm using genetic approach

This paper proposes a robust and fully automated respiratory phase segmentation method using single channel tracheal breath sounds (TBS) recordings of different types. The estimated number of respiratory segments in a TBS signal is firstly obtained based on noise estimation and nonlinear mapping. Respiratory phase boundaries are then located through the generations of multi-population genetic algorithm by introducing a new evaluation function based on sample entropy (SampEn) and a heterogeneity measure. The performance of the proposed method is analyzed for single channel TBS recordings of various types. An overall respiratory phase segmentation accuracy is found to be 12 ± 5 ms for normal TBS and 21 ± 9 ms for adventitious sounds. The results show the robustness and effectiveness of the proposed segmentation method. The proposed method has been a successful attempt to solve the clinical application challenge faced by the existing phase segmentation methods in terms of respiratory dysfunctions.     

Prostate segmentation algorithm using dyadic wavelet transform and discrete dynamic contour

Knowing the location and the volume of the prostate is important for ultrasound-guided prostate brachytherapy, a commonly used prostate cancer treatment method. The prostate boundary must be segmented before a dose plan can be obtained. However, manual segmentation is arduous and time consuming. This paper introduces a semi-automatic segmentation algorithm based on the dyadic wavelet transform (DWT) and the discrete dynamic contour (DDC). A spline interpolation method is used to determine the initial contour based on four user-defined initial points. The DDC model then refines the initial contour based on the approximate coefficients and the wavelet coefficients generated using the DWT. The DDC model is executed under two settings. The coefficients used in these two settings are derived using smoothing functions with different sizes. A selection rule is used to choose the best contour based on the contours produced in these two settings. The accuracy of the final contour produced by the proposed algorithm is evaluated by comparing it with the manual contour outlined by an expert observer. A total of 114 2D TRUS images taken for six different patients scheduled for brachytherapy were segmented using the proposed algorithm. The average difference between the contour segmented using the proposed algorithm and the manually outlined contour is less than 3 pixels.     

Masseter segmentation using an improved watershed algorithm with unsupervised classification

The watershed algorithm always produces a complete division of the image. However, it is susceptible to over-segmentation and sensitivity to false edges. In medical images this leads to unfavorable representations of the anatomy. We address these drawbacks by introducing automated thresholding and post-segmentation merging. The automated thresholding step is based on the histogram of the gradient magnitude map while post-segmentation merging is based on a criterion which measures the similarity in intensity values between two neighboring partitions. Our improved watershed algorithm is able to merge more than 90% of the initial partitions, which indicates that a large amount of over-segmentation has been reduced. To further improve the segmentation results, we make use of K-means clustering to provide an initial coarse segmentation of the highly textured image before the improved watershed algorithm is applied to it. When applied to the segmentation of the masseter from 60 magnetic resonance images of 10 subjects, the proposed algorithm achieved an overlap index (κ) of 90.6%, and was able to merge 98% of the initial partitions on average. The segmentation results are comparable to those obtained using the gradient vector flow snake.     

Morphometry of spermatozoa using semiautomatic image analysis.

Human sperm heads were detected and tracked using semiautomatic image analysis. Measurements of size and shape on two specimens from each of 26 men showed that the major component of variability both within and between subjects was the number of small elongated sperm heads. Variability of the computed features between subjects was greater than that between samples from the same subject.     

A dose-volume histogram based optimization algorithm for ultrasound guided prostate implants

The task of treatment planning for prostate implants is to find an optimal seed configuration, comprising the target coverage and dosimetric consideration of critical structures such as the rectum and urethra. An efficient method to accomplish this is to use an inverse planning technique that derives the optimized solution from a prescribed treatment goal. The goal can be specified in the voxel domain as the desired doses to the voxels of the target and critical structures, or in the dose volume representation as the desired dose volume histograms (DVHs) of the target and critical structures. The DVH based optimization has been successfully used in plan optimization for intensity-modulated radiation therapy (IMRT) but little attention has been paid to its application in prostate implants. Clinically, it has long been known that some normal structure tolerances are more accurately assessed by volumetric information. Dose-volume histograms are also widely used for plan evaluation. When working in the DVH domain for optimization one has more control over the final DVHs. We have constructed an objective function sensitive to the DVHs of the target and critical structures. The objective function is minimized using an iterative algorithm, starting from a randomly selected initial seed configuration. At each iteration step, a trial position is given to a randomly selected source and the trial position is accepted if the objective function is decreased. To avoid being trapped in a less optimal local minimum, the optimization process is repeated. The final plan is selected from a pool of optimized plans obtained from a series of randomized initial seed configurations.     

Testing the effects of educational strategies on comprehension of a genomic concept using virtual reality technology

Objective

Applying genetic susceptibility information to improve health will likely require educating patients about abstract concepts, for which there is little existing research. This experimental study examined the effect of learning mode on comprehension of a genomic concept.

Methods

156 individuals aged 18–40 without specialized knowledge were randomly assigned to either a virtual reality active learning or didactic learning condition. The outcome was comprehension (recall, transfer, mental models).

Results

Change in recall was greater for didactic learning than for active learning (p < 0.001). Mean transfer and change in mental models were also higher for didactic learning (p < 0.0001 and p < 0.05, respectively). Believability was higher for didactic learning (p < 0.05), while ratings for motivation (p < 0.05), interest (p < 0.0001), and enjoyment (p < 0.0001) were higher for active learning, but these variables did not mediate the association between learning mode and comprehension.

Conclusion

These results show that learning mode affects comprehension, but additional research is needed regarding how and in what contexts different approaches are best for educating patients about abstract concepts.

Practice implications

Didactic, interpersonal health education approaches may be more effective than interactive games in educating patients about abstract, unfamiliar concepts. These findings indicate the importance of traditional health education approaches in emerging areas like genomics.