Observing deep radiomics for the classification of glioma grades

Abstract

Deep learning is a promising method for medical image analysis because it can automatically acquire meaningful representations from raw data. However, a technical challenge lies in the difficulty of determining which types of internal representation are associated with a specific task, because feature vectors can vary dynamically according to individual inputs. Here, based on the magnetic resonance imaging (MRI) of gliomas, we propose a novel method to extract a shareable set of feature vectors that encode various parts in tumor imaging phenotypes. By applying vector quantization to latent representations, features extracted by an encoder are replaced with a fixed set of feature vectors. Hence, the set of feature vectors can be used in downstream tasks as imaging markers, which we call deep radiomics. Using deep radiomics, a classifier is established using logistic regression to predict the glioma grade with 90% accuracy. We also devise an algorithm to visualize the image region encoded by each feature vector, and demonstrate that the classification model preferentially relies on feature vectors associated with the presence or absence of contrast enhancement in tumor regions. Our proposal provides a data-driven approach to enhance the understanding of the imaging appearance of gliomas.

Effects of Multi-Shell Free Water Correction on Glioma Characterization

“…Necrosis can be detected based on expert evaluations of MRI (post-contrast T 1 w images in particular), but quantitative thresholds of FW content to characterize necrotic regions are not routinely applied in the clinic. Some studies have suggested that a more accurate description of the necrotic region helps with predicting survival and tumor aggressiveness in glioblastoma patients [ 40 , 41 ], and deep learning-based segmentation [ 42 ] and tumor classification [ 43 ] are being explored. Adding the estimation of FW to such analyses is a possible, clinically useful application of FW estimation.…”

Machine Learning-Based Analysis of Glioma Grades Reveals Co-Enrichment

“…However, the subtyping system update did not influence the grading system that is based on the histological criteria derived from a biological behavior of neoplasm. Specifically, WHO discerns four glioma grades that are defined as follows: grade I (GI or G1) with low proliferative potential, grade II (GII or G2) with low-level proliferative activity, grade III (GIII or G3) histological evidence of malignancy and grade IV (GIV or G4) cytologically malignant that is the most malignant form of glioma [ 5 ]. Here, the grading system was adopted from The Cancer Genome Atlas (TCGA) which classifies gliomas into lower-grade gliomas (LGG) including GII and GIII [ 6 ] and glioblastoma multiforme (GBM) including GIV.…”

Application of Artificial Intelligence in COVID-19 Diagnosis and Therapeutics

“…The medical field is no exception, and many AI-powered medical devices have been developed and are now being applied clinically [ 15 ]. In medical research, AI has been introduced into various tasks, including medical image analysis, such as radiological image analysis, endoscopic image analysis, and pathological image association analysis; omics analysis such as genome analysis, epigenome analysis, and proteome analysis; and natural language processing for drug discovery, electronic medical record information, and literature search [ 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ]. Importantly, research on COVID-19 is being conducted worldwide, and AI is now being actively used in vaccine development, the development of new diagnostic methods, and the development of new therapeutic agents by extracting important features from vast amounts of data.…”