Explore chapters and articles related to this topic
Basic Approaches of Artificial Intelligence and Machine Learning in Thermal Image Processing
Published in U. Snekhalatha, K. Palani Thanaraj, Kurt Ammer, Artificial Intelligence-Based Infrared Thermal Image Processing and Its Applications, 2023
U. Snekhalatha, K. Palani Thanaraj, Kurt Ammer
The GLCM is based on a statistical method that is used for analyzing the texture feature which gives information about the spatial relationship of pixels. In an image, both the column and row will be identical to the number of gray levels denoted as “g” in the GLCM matrix. The texture of the image is used to calculate the pixel with particular values in a specific spatial relationship. Data are divided into first, second, and higher-order statistics. In that way, the second order is a statistical method in the gray level co-occurrence matrix.
Radiogenomics
Published in Ruijiang Li, Lei Xing, Sandy Napel, Daniel L. Rubin, Radiomics and Radiogenomics, 2019
In the case of pre-defined image features, current implementations of quantitative imaging pipelines evolve rapidly with increasingly more features being extracted from medical images [52,53]. Each class of features is defined by a specific characteristic that the class of features represents. Roughly, quantitative image features can be grouped in to four different areas: edge, shape, intensity, and texture [52]. Edge and shape features are more straightforward as they represent quantitative characteristics of the tumor edge, e.g., sharp versus blurry edges. Shape features represent the overall shape of the tumor including how spherical it is and how irregular the boundary is defined. For example, the compactness features are defined as follows: a value of 1.0 reflects a lesion that is completely spherical, whereas a value of 0 reflects a completely irregular lesion. Next, other features that capture shape are, for example, the radial distance signal (RDS), this is a set of features that captures the variability of the tumor shape [52]. Intensity and texture features represent first-, second-, and higher-order of statistics of the image and become increasingly harder to interpret. First-order statistics involve histogram statistics such as mean, median, skewness, and kurtosis of the intensity histogram. These features do not take into account the spatial organization of each voxel in the tumor, but rather only capture a summary of the intensity histogram of a tumor. Beyond that, higher-order statistics become increasingly less interpretable, but do capture spatial information. These are the so-called texture features and are often the most important contribution of the radiomics characterization. There is an increasingly growing list of texture features and more are being defined. Texture features were first introduced by Haralick [54]. The first texture features describe spatial relationships between voxels through a gray-level co-occurrence matrix (GLCM) defined as combinations of discretized gray levels of neighboring voxels and how they are distributed in different directions. To make GLCM features rotation invariant, these features are calculated after combining information from the different matrices [55]. Similar to GLCM, the gray-level run-length matrix (GLRLM) was introduced in 1975 [56] to capture a run length defined as the length of a consecutive sequence of pixels or voxels with the same gray level along a particular direction. Subsequently other texture features have been introduced such as the gray-level size-zone matrix (GLSZM) [57], gray-level distance-zone matrix (GLDZM) [57], and neighborhood gray-tone difference matrix (NGTDM) [58]. Taken together, there is a potential to extract 100s–1000s of texture features out of a single radiology image. However, it is well known that correlation between texture features can be high, and as such, not all features represent unique characteristics of the MR image, rather this approach is motivated by extracting any useful signal out of the image. In a second step, the radiomics research will use machine learning approaches to analyze the data and identify which features are important either using unsupervised or supervised analysis.
Artificial intelligence: improving the efficiency of cardiovascular imaging
Published in Expert Review of Medical Devices, 2020
Andrew Lin, Márton Kolossváry, Ivana Išgum, Pál Maurovich-Horvat, Piotr J Slomka, Damini Dey
The ability of ML techniques to handle high-dimensional data has facilitated the growth of a new field in medical imaging known as radiomics. This refers to the process of extracting a large number of quantitative imaging features from a given region of interest to create big data in which each abnormality is characterized by hundreds of parameters extending far beyond those that can be characterized by the human eye [103]. The quantitative features are calculated using dedicated software, which accepts the image datasets as inputs. Following feature extraction, datamining and ML approaches are used to find new, meaningful patterns between the different parameters, with the aim of identifying novel imaging biomarkers which may reflect the underlying pathophysiology of a tissue. Radiomic features can be typically be classified as shape-based, intensity-based, or texture-based [103]. Shape features describe the geometric properties of the region of interest, while intensity-based metrics are calculated from the absolute voxel values themselves. Texture analysis is the modeling of the spatial distribution of voxel gray-level intensities, using second- or higher-order statistics to provide a measure of heterogeneity within the region of interest [104].
An updated review on the diagnosis and assessment of post-treatment relapse in brain metastases using PET
Published in Expert Review of Neurotherapeutics, 2022
Norbert Galldiks, Michael Wollring, Jan-Michael Werner, Michel Friedrich, Gereon R. Fink, Karl-Josef Langen, Philipp Lohmann
After these preprocessing steps, radiomics features can be extracted. These features have the potential to uncover tumoral characteristics that are beyond the means of human perception. Basically, shape features (i.e. geometrical properties), histogram-based features (i.e. distribution of individual voxel intensity values), textural features (i.e. statistical relationships between the intensity values of neighboring voxels and groups of voxels), and higher-order statistics features (i.e. features extracted after the application of mathematical transformations such as filters) can be extracted. Thus, hundreds to thousands of features can be easily obtained from the respective medical image modality.
Epilepsy classification using optimized artificial neural network
Published in Neurological Research, 2018
Jagriti Saini, Maitreyee Dutta
At this stage 15 unique features based on he first-order, second-order, and higher-order statistics are extracted from all 300 signals. The feature extraction process is performed using MATLAB 2015. All obtained parameters are normalized using the mathematical formula of the Min-Max Approach, so that they can fit inside a standard range of 0 and 1.