This paper describes several grayscale image processing and segmentation algorithms that play important roles in medical image analysis. First, we describe grayscale image processing techniques including smoothing and local feature extraction. Second, we introduce several image segmentation frameworks including active contour models such as snakes and level set, graph cuts, and atlas-based method. We also refer to the techniques to incorporate some anatomical knowledge into each of the segmentation methods.
The term “registration” has been used to refer to establishing pixel correspondence between two images in medical image analysis. In computer-integrated surgical system such as surgical navigation and surgical robot, it refers to compute relative transformation between coordinate systems such as a local coordinate system defined in patient space and a global coordinate system defined in the operating room. Both applications share the same theoretical background except that the registration of image (i.e., a cluster of intensity values on the integer lattice) involves interpolation, because the transformed point does not lie exactly on the integer lattice. Due to the difference in the application area, the two registrations are rarely explained with a unified terminology. In this article, we overview the common elemental tools in registration, i.e., transformation function, similarity function and optimization method, with the emphasis on the comparison between the two applications, and we also describe the future perspective of the research field.
In this paper, a brief overview of binary images and binary image processing is introduced. Generally, medical images and intraoperative movies are gray images or color images, and binary images are obtained by thresholding these images. The purposes of the binary image processing are to extract regions of interest and to analyze the distribution or shape of those regions.
This article reviews feature descriptors for image recognition tasks. It ranges from basic statistics of images, filters, local descriptors, and bag-of-words representations and the use of convolutional neural network.
Advancement of medical diagnosis device and computing technology leads the establishment of image guided therapy which realize precise guidance of treatment procedure using diagnosis images and medical image processing technologies. In this paper, I introduce the properties of these image modalities and technologies in the field of interventional radiology and minimally invasive surgery, including image enhancement and segmentation, registration, and organ motion estimation and tracking. On this technologies, we should define the requirement of accuracy and processing time for each treatment properly, based on clinical and engineering knowledge.
Alzheimer's disease is the most cases in dementia, and a diagnosis for its very early stage is required. A deposit of amyloid beta (Aβ) is recognized to cause Alzheimer's disease, and it can be visualized using PET, an amyloid imaging. An amyloid imaging therefore promises a useful tool for Alzheimer's diagnosis. A highly quantitative measurement of Aβ deposit is required for an effective diagnosis to Alzheimer's disease with PET, and thus, this article aims to survey some aspects in an amyloid imaging: the theory to quantify amyloid deposit using PET, a necessity of a dynamic PET study that acquired a time history of radioactivity in tissues with multiple PET scans, and a method of early images to reduce an issue of a total PET scan time.
Blood vessels in retinal fundus image are the only vessels which we can observe directly from outside human body. Therefore, funduscopy is employed in the complete physical examination for the early detection of hypertension and arteriosclerosis. With the increase of the number of examinees in recent years, it is increasing the burden of the doctor's image interpretations. To support their image interpretations, several computer-aided diagnosis (CAD) schemes have been developed. In the CAD schemes, vessel extraction method in retinal fundus images is an essential component. In this paper, we proposed a new method for extracting blood vessels in retinal fundus images, which combines morphology filter bank and AdaBoost. The morphology filter bank was used for extracting line patterns in the determination of the initial candidates of blood vessels. The AdaBoost is employed for eliminating false positives in the initial candidates. We evaluated our proposed method by using training and testing datasets in DRIVE database. The result indicated that sensitivity was 0.7362 and specificity was 0.9714. Our propose method is useful for the extraction of vessels in retinal fundus images.
Detailed knowledge of the relation between tissue-specific acoustic properties and histologic features is essential to achieve highly accurate quantitative ultrasonography results. This study evaluated the relation between acoustic properties and microscopic histologic features, especially focusing on acoustic impedance and speed of sound in rat livers of four histologic types: healthy livers, fatty livers, livers with non-alcoholic steatohepatitis (NASH), fibrotic livers. An 80-MHz center frequency transducer (resolution 20 μm) equipped with a scanning acoustic microscopy system was employed. Statistical analysis showed that both acoustic impedance and speed of sound were slightly lower in lipid-rich tissue (i.e., fatty and NASH liver) than in healthy rat liver and were both higher in fibrotic liver than in the other types (p<0.01). This tendency suggests that these two parameters (impedance, speed of sound) indicated the presence of tissue degeneration caused by lipid deposits or fibrosis. It is thus possible to use them to distinguish an unhealthy from a healthy liver.
In our previous paper (J Dong, H Kudo: Proposal of compressed sensing using nonlinear sparsifying transform for CT image reconstruction. Medical Imaging Technology. Vol. 34 pp235-244, 2016), we showed that nonlinear sparsifying transform provides a new framework of compressed sensing (CS) for sparse-view CT image reconstruction. Furthermore, it was experimentally demonstrated to have superiority in improving image quality compared to total variation (TV) minimization, which is the most standard approach in CS. The image quality improvement appears in removing patchy artifacts, preserving accurate object boundaries, and preserving image textures. The TV uses the gradient transform which considers only correlations between adjacent pixels, while the nonlinear sparsifying transform can consider evaluating intensity variations among a specified relatively large search window. This property can be considered to be the key reason for achieving image quality improvement by the nonlinear sparsifying transform. However, the iterative algorithm developed in our previous paper, which can be viewed as a special case of standard iterative-thresholding (IT) algorithm, suffers from a drawback that it converges very slowly leading to a long computation time. The main reason of slow convergence is that the IT algorithm belongs to a class of simultaneous iterative algorithms, in which all projection data are used simultaneously (in parallel) for each image update. However, as is well-known in past research activities of CT image reconstruction, it is expected that the convergence can be significantly accelerated by introducing a class of row-action or block-iterative algorithm. Based on this observation, in this paper, we propose an accelerated algorithm of the CS using nonlinear sparsifying transform. By using proximal splitting framework, we succeeded in performing image update with a row-action-type program. The row-action-type update showed an encouraging acceleration such that both the iteration number and the computation time were reduced significantly compared to our previous simultaneous iterative algorithm. We investigated the efficiency of proposed accelerated algorithm using a numerical phantom and a practical CT image.