Automatic Segmentation of Polyps in Endoscopic Image Using Level-Set Formulation

Due to the low contrast, blurry boundaries, and large amount of shadows in breast ultrasound (BUS) images, automatic tumor segmentation remains a challenging task. Deep learning provides a solution to this problem, since it can effectively extract representative features from lesions and the background in BUS images. A novel automatic tumor segmentation method is proposed by combining a dilated fully convolutional network (DFCN) with a phase-based active contour (PBAC) model. The DFCN is an improved fully convolutional neural network with dilated convolution in deeper layers, fewer parameters, and batch normalization techniques; and has a large receptive field that can separate tumors from background. The predictions made by the DFCN are relatively rough due to blurry boundaries and variations in tumor sizes; thus, the PBAC model, which adds both region-based and phase-based energy functions, is applied to further improve segmentation results. The DFCN model is trained and tested in dataset 1 which contains 570BUS images from 89 patients. In dataset 2, a 10-fold support vector machine (SVM) classifier is employed to verify the diagnostic ability using 460 features extracted from the segmentation results of the proposed method. Advantages of the present method were compared with three state-of-the-art networks; the FCN-8s, U-net, and dilated residual network (DRN). Experimental results from 170BUS images show that the proposed method had a Dice Similarity coefficient of 88.97±10.01%, a Hausdorff distance (HD) of 35.54±29.70 pixels, and a mean absolute deviation (MAD) of 7.67±6.67 pixels, which showed the best segmentation performance. In dataset 2, the area under curve (AUC) of the 10-fold SVM classifier was 0.795 which is similar to the classification using the manual segmentation results. The proposed automatic method may be sufficiently accurate, robust, and efficient for medical ultrasound applications.

One key task in forensic science is to perform criminal investigation through image database retrieval. Of the various images, tire pattern is an important type of image data for crime scene investigation. However, different rotation and direction of tire patterns are often encountered and is insufficient to use the conventional multi-scale texture feature extraction method which is not rotational invariant. To alleviate this problem, the paper proposed two new texture feature extraction methods based on the Radon transform and Curvelet transform. The experiments were conducted using a tire pattern database containing 400 images. The results show that the proposed methods effectively overcome the influences of rotation and significantly improve the retrieval efficiency.

Matching-constrained active contours with affine-invariant shape prior

Automatic bone segmentation of Computed Tomography (CT) images is an important step in imageguided surgery where segmentation errors could be critical. Previous attempts include intensity-, edge-, region-, and deformable curve-based approaches [1], but none claims fully satisfactory performance. In this study, we have tested most widely used active contour (AC) -based approaches to segment knee bones from CT imagery, namely the Gradient Vector Flow (GVF) AC, the original geometric AC, the geodesic AC, the GVF fast geometric AC, and the Chan-Vese multiphase AC without edges. Among the techniques, only the Chan-Vese multiphase AC demonstrated satisfactory performance, proving its suitability for bone segmentation from CT images.

Active contours driven by divergence of gradient vector flow

Automatic segmentation and tracking of biological objects from dynamic microscopy data is of great interest for quantitative biology. A successful framework for this task are active contours, curves that iteratively minimize a cost function, which contains both data-attachment terms and regularization constraints reflecting prior knowledge on the contour geometry. However the choice of these latter terms and of their weights is largely arbitrary, thus requiring time-consuming empirical parameter tuning and leading to sub-optimal results. Here, we report on a first attempt to use regularization terms based on known biophysical properties of cellular membranes. The present study is restricted to 2D images and cells with a simple to skeletal cortex underlying the membrane. We describe our new active contour model and its implementation, and show a first application to real biological images. The obtained segmentation is slightly better than standard active contours, however the main advantage lies in the self-consistent and automated determination of the weights of regularization terms. This encouraging result will lead us to extend the approach to 3D and more complex cells.

Colorectal cancer (CRC) is a leading cause of cancer deaths worldwide. Although polypectomy at early stage reduces CRC incidence, 90% of the polyps are small and diminutive, where removal of them poses risks to patients that may outweigh the benefits. Correctly detecting and predicting polyp type during colonoscopy allows endoscopists to resect and discard the tissue without submitting it for histology, saving time, and costs. Nevertheless, human visual observation of early stage polyps varies. Therefore, this paper aims at developing a fully automatic algorithm to detect and classify hyperplastic and adenomatous colorectal polyps. Adenomatous polyps should be removed, whereas distal diminutive hyperplastic polyps are considered clinically insignificant and may be left in situ . A novel transfer learning application is proposed utilizing features learned from big nonmedical datasets with 1.4-2.5 million images using deep convolutional neural network. The endoscopic images we collected for experiment were taken under random lighting conditions, zooming and optical magnification, including 1104 endoscopic nonpolyp images taken under both white-light and narrowband imaging (NBI) endoscopy and 826 NBI endoscopic polyp images, of which 263 images were hyperplasia and 563 were adenoma as confirmed by histology. The proposed method identified polyp images from nonpolyp images in the beginning followed by predicting the polyp histology. When compared with visual inspection by endoscopists, the results of this study show that the proposed method has similar precision (87.3% versus 86.4%) but a higher recall rate (87.6% versus 77.0%) and a higher accuracy (85.9% versus 74.3%). In conclusion, automatic algorithms can assist endoscopists in identifying polyps that are adenomatous but have been incorrectly judged as hyperplasia and, therefore, enable timely resection of these polyps at an early stage before they develop into invasive cancer.

We present an efficient algorithm for the robust segmentation of vessel contours in Computed Tomography Angiography (CTA) images. The algorithm performs its task within several steps based on a 3D Active Contour Model (ACM) with refinements on Multi-Planar Reconstructions (MPRs) using 2D ACMs. To be able to distinguish true vessel edges from spurious, an adaptive self-learning edge model is applied. We present details of the algorithm together with an evaluation on n=150 CTA data sets and compare the results of the automatic segmentation with manually outlined contours resulting in a median dice similarity coefficient (DSC) of 92.2%. The algorithm is able to render 100 contours within 1.1s on a Pentium®4 CPU 3.20 GHz, 2 GByte of RAM.

The endoscopic images do not contain sharp edges to segment using the traditional segmentation methods for obtaining edges. Therefore, the active contours or 'snakes' using level set method with the energy minimization algorithm is adapted here to segment these images. The result obtained from the above segmentation process will be of number of segmented regions. The analysis of the segmentation results shows that the abnormal images generate more number of regions as compared to the normal images.

This work aims to develop a robust and efficient method to track the fuzzy borders between liver and the abutted organs where automatic liver segmentation usually suffers, and to investigate its applications in automatic liver segmentation on noncontrast-enhanced planning computed tomography (CT) images. In order to track the fuzzy liver-chestwall and liver-heart borders where oversegmentation is often found, a starting point and an ending point were first identified on the coronal view images; the fuzzy border was then determined as a geodesic curve constructed by minimizing the gradient-weighted path length between these two points near the fuzzy border. The minimization of path length was numerically solved by fast-marching method. The resultant fuzzy borders were incorporated into the authors' automatic segmentation scheme, in which the liver was initially estimated by a patient-specific adaptive thresholding and then refined by a geodesic active contour model. By using planning CT images of 15 liver patients treated with stereotactic body radiation therapy, the liver contours extracted by the proposed computerized scheme were compared with those manually delineated by a radiation oncologist. The proposed automatic liver segmentation method yielded an average Dice similarity coefficient of 0.930 ± 0.015, whereas it was 0.912 ± 0.020 if the fuzzy border tracking was not used. The application of fuzzy border tracking was found to significantly improve the segmentation performance. The mean liver volume obtained by the proposed method was 1727 cm(3), whereas it was 1719 cm(3) for manual-outlined volumes. The computer-generated liver volumes achieved excellent agreement with manual-outlined volumes with correlation coefficient of 0.98. The proposed method was shown to provide accurate segmentation for liver in the planning CT images where contrast agent is not applied. The authors' results also clearly demonstrated that the application of tracking the fuzzy borders could significantly reduce contour leakage during active contour evolution.

Data-driven prognostic methods typically make use of observer signals reflective of the system health combined with machine learning methods to predict the Remaining Useful Life (RUL) of the system. Currently, majority of feature extraction methods developed for prognostics focused on extracting features from regular time series applications. However, events-data collected during occurrence of an event are stochastic in nature with irregular sampling frequency, which is challenging for current methods. For most prognostic applications, the RUL is closely correlated with changes in data trend exhibited in the observer signals. Motivated by this phenomenon, this paper proposes a novel Time Series-Histogram of Features method, which extracts features describing the local degradation features exhibited by observer signals in a moving time window. The proposed method is illustrated via a case study on a benchmark simulated aero-engine dataset. Results indicate that the proposed methodology performs as well as or better than conventional feature extraction methods on the same time window of information. Furthermore, it is also shown that the proposed method extracts information complementary to conventional feature extraction techniques, thus resulting in superior performance by combining these feature extraction techniques.

Automotive industries in the last decade are demanding for lightweight, corrosion resistance and improved fatigue performance materials. Composites having all such properties quickly gain popularity. Due to complex fabrication methods, composites are known for various process defects. Thus, monitoring and characterizing the composite component before assembly is necessary to maintain the overall structural integrity. Further a rapid and reliable inspection is much needed addition in most automotive industries to save time.
 Non-Destructive Evaluation (NDE) is popular for component testing in the automobile industry because they do not cause any permanent alteration to components. Some NDE techniques require sensors (Piezo Electric transducers, PVDF films, Optical fiber, etc.) that need to be bonded to the structure for testing. Some of these sensors are expensive and cannot be reused once detached. This work aims at the reusability of sensors. Further, the reusable sensors can be deployed in an array configuration for multi-purpose NDE. SMART Skin shall be explained as a Multiple-Transmitter-Multiple-Receiver (MTMR) Piezo-ceramic based sensor (PZT) array which is embedded to a conformable skin. The bottom layer of the skin is coated with pressure-sensitive adhesive to be attached to most curved and non-curved structural surfaces. Each PZT sensor nodes are individually controlled by a MATLAB code that actuates and receive the GW waves signals.
 In this work, we have validated the ability of rapid material characterization using isotropic materials.
 Using experimental data collected by the proposed methodology, we have developed an automated material classification algorithm using various machine learning algorithms. A novel application of functional data analysis (fDA) which converts discrete samples into continuous curves is used. The curves are represented as linear combinations of basis functions. The advantage of this fDA method is that we use the shape of the signal instead of extracting features from the signal. Several choices of basis coefficients such as fourier series, wavelet and B-spline are considered of which spline basis gives the best results. These basis coefficients are then feed into machine learning models like Support Vector Machine (SVM), Random Forest (RF), k nearest neighbor (knn) for classification. The efficacy of this novel method is compared with conventional feature extraction techniques such as zero-crossing coefficients, absolute maximum value, different statistical features like mean and variance, energy of the signal. Encouraging results are obtained that shows the fDA methodology is efficient over the conventional feature extraction methods as it improves the prediction performance on the classifier and result in a faster and cost-effective model by reducing the predictor dimensionality.

Pulmonary diseases are the cause of millions of deaths around the world. Chest radiography is one of the most used medical imaging modality for the identification and diagnosis of pulmonary abnormalities. However, segmentation of the pulmonary region is important to obtain objective information and extract characteristics necessary for the diagnosis. In the literature are available some pulmonary segmentation methods based on different processing techniques. However, these still have some limitations. Consequently, we present an automatic pulmonary segmentation approach, based on Enhancement by Wavelet, Mathematical Morphology and Active Contour. The methodology is divided into three steps. First, the original image is enhanced by discrete wavelet. Next, occurs a combination of the Otsu threshold followed by a series of morphological operations to identify the pulmonary object; hence, pulmonary tissue information is discriminated and binarized. Finally, Active Contour method improves the binary information of pulmonary tissue, mainly in the border region. The evaluation was performed in 247 digital chest X-ray images of human. The high accuracy and robustness of the proposed method was demonstrated with values of Overlap (OR (%)) = 95.9 ± 2.9, and Average Contour Distance (ACD (mm)) = 0.76 ± 0.92, better than currently literature results.

This paper presents an automatic tissue segmentation methodology for High-Resolution Ultrasonic Transmission Tomography (HUTT) imagery of biological organs. This method combines a recent segmentation approach: the L-level set active contours algorithm with unsupervised clustering using the agglomerative hierarchical k-means algorithm. The active contours algorithm has been recently explored as a powerful tool for image segmentation since it automatically decomposes a given image into 2L segment classes by utilizing L level set functions and finding the optimal boundaries of the 2L segment classes so that the pixel feature values of each segment are as homogeneous as possible. Unfortunately, the algorithm is often trapped at local minima due to the intrinsic non-convexity of the cost function, especially for noisy data. To overcome this problem, we introduce a multi-stage multi-resolution analysis that optimizes the active contours at successive resolutions of the image data. The resulting segments are then re-clustered by subsequent agglomerative hierarchical k-means clustering that seeks the optimal clusters yielding the minimum within-cluster distance in the feature space. The preliminary studies reported here indicate that this proposed methodology can enhance the accuracy of soft tissue segmentation and provide fully automatic tissue differentiation without any user intervention except for specifying the number of level set functions L.

A segmentation framework for abdominal organs from CT scans