Precision Imaging of Prostate Cancer
Ayman El-Baz, Gyan Pareek, Jasjit S. Suri in Prostate Cancer Imaging, 2018
Image registration is a process of aligning two or more images, which aims to find the optimal transformation that best aligns the structures of interest in the input images. Image registration is needed in order to integrate the features from different images of mpMRI such as DCE-MRI and T2W MRI. The registration of images requires the selection of the feature space, a similarity measure, a transformation type, and a search strategy [106]. The DICOM header of MR images can provide coordination and orientation information that are useful for registering T2W, ADC, and Ktrans maps. T2W-MRI is considered as the reference. Other modalities can be registered to T2W-MRI by aligning the coordinates of their origins, which are obtained from the DICOM header. If necessary, resolution adjustment is also performed after the alignment.
A Combination of Dilated Adversarial Convolutional Neural Network and Guided Active Contour Model for Left Ventricle Segmentation
Kayvan Najarian, Delaram Kahrobaei, Enrique Domínguez, Reza Soroushmehr in Artificial Intelligence in Healthcare and Medicine, 2022
In this class of segmentation techniques, an atlas that describes the structure of the target object is first generated from one or more manually delineated images. Image registration is a process of mapping the coordinates of one image to those of another image. Given an atlas, a new image can be segmented by image registration between the new image and the atlas. A three-step registration framework was proposed in Zhuang et al. (2010). A global affine registration was first applied to localize the heart, then a locally affine registration method was employed to initialize the substructures. Finally, a free-form deformation with adaptive control point status was proposed to refine the local details. The method achieved good performance on heart images with diverse morphology and pathology. In Lorenzo-Valdés et al. (2004), a 4D probabilistic heart atlas was constructed from 14 healthy cases, which encodes both spatial anatomical information and temporal information. With the 4D probabilistic atlas, the expectation-maximization algorithm was applied to perform the segmentation.
Multimodality image fusion
Yi-Hwa Liu, Albert J. Sinusas in Hybrid Imaging in Cardiovascular Medicine, 2017
From the technical point of view, in the majority of studies, image registration is performed by means of surface-based algorithms: the original images are preprocessed to extract landmarks that are eventually used for the final alignment. In the last few decades, nuclear cardiac imaging has been the focus of extensive research in order to objectively and automatically quantify perfusion images (left ventricle [LV] epicardial and endocardial boundaries, wall motion, ejection fraction, etc.). A number of software packages have been developed and made available to clinicians that allowed the standardization and the widespread utilization of nuclear cardiology as the most reliable tool for CAD assessment. Image processing of anatomical images on the other hand has lagged behind with regard to automated feature extraction.
A comparison of magnetic resonance imaging techniques used to secure biopsies in prostate cancer patients
Published in Expert Review of Anticancer Therapy, 2019
Annemarijke van Luijtelaar, Joyce Bomers, Jurgen Fütterer
MRI-TRUS fusion offers the benefits of cognitive fusion biopsy and allows clinicians to both visualize and target specific suspected lesions within the prostate under software guidance. Different software-based fusion platforms have been developed to help ensure the fusion of the mpMRI with the real-time US, each with its own specific features. The main difference between the number of commercially available software-based fusion platforms is the image registration methods being either rigid or non-rigid (elastic). Rigid image registration occurs by aligning the mpMRI onto the TRUS images while not accounting for movements by the patient and without adjustment for possible deformation of the prostate by the introduction of the TRUS probe. Non-rigid image registration matches corresponding point landmarks and happens during the procedure by accounting for any real-time changes of the prostate and patients movements [39]. As non-rigid image registration is compensating for the deformation during the biopsy, it is likely that non-rigid image registration would be more accurate compared to rigid image registration. However, studies have shown similar detection rates for csPCa between rigid and elastic image registration for MRI-TRUS fusion-guided biopsy [44]. Furthermore, the software-based fusion platforms differ in the method of needle tracking, software functionalities and the biopsy route (transrectal or transperineal). Table 4 displays an overview of the commonly used software-based fusion platforms.
Primary central nervous system lymphoma and glioblastoma differentiation based on conventional magnetic resonance imaging by high-throughput SIFT features
Published in International Journal of Neuroscience, 2018
Yinsheng Chen, Zeju Li, Guoqing Wu, Jinhua Yu, Yuanyuan Wang, Xiaofei Lv, Xue Ju, Zhongping Chen
Different from normal radiomics studies which calculate high throughput image features from several aspects usually including intensity, shape, texture and wavelet, we proposed to calculate the radiomics features by using scale invariant feature transform (SIFT) [20]. As a famous algorithm for point feature extraction, since its proposition by Lowe [20], SIFT has been successfully utilized in image registration [21], object recognition [22,23], video tracking [24] and so on. The basic idea of SIFT is to find the keypoints of an image by using difference-of-Gaussian operation and local extreme detection across scale spaces. The local features of keypoints are then obtained by calculating the gradient magnitude and orientation of each keypoint. Since the SIFT feature is invariant to image scaling and rotation, insensitive to image intensity and noise, it is ideal to describe local voxel arrangement of MRI.
Towards a generalised development of synthetic CT images and assessment of their dosimetric accuracy
Published in Acta Oncologica, 2020
Josefine Handrack, Mark Bangert, Christian Möhler, Tilman Bostel, Steffen Greilich
The usage of BDA specifically for pelvic cancer has been widely investigated in the literature, however mostly for water equivalent (WE) bulk densities [4,17] or WE and bone [4,5,18] as surrogate EDs. The commercial approach described in Köhler et al. [14] increased the complexity of the sCT by using five different tissue classes for BDA. The next increase in complexity is the dual model [19,20], which separates bone (polynomial fit) and soft tissue (interpolation between fat, muscle and urine) for sCT generation. Machine learning and image registration are the most complex and computational demanding approaches. However, at some point an increase in complexity of the sCT generation might not result in a relevant increase in dosimetric accuracy any more. This trade-off has not yet been analysed systematically.
Related Knowledge Centers
- Positron Emission Tomography
- Ultrasound
- Prostate
- Medical Imaging
- Statistical Parametric Mapping
- Air
- Computational Anatomy
- Large Deformation Diffeomorphic Metric Mapping
- CT Scan
- Magnetic Resonance Imaging