Handbook of Medical Image Computing and Computer Assisted Intervention

Handbook of Medical Image Computing and Computer Assisted Intervention

The Elsevier and MICCAI Society Book Series
2020, Pages 429-456
Handbook of Medical Image Computing and Computer Assisted Intervention

Chapter 18 - Radiomics: Data mining using quantitative medical image features

https://doi.org/10.1016/B978-0-12-816176-0.00023-5Get rights and content

Abstract

Radiomics uses multiple image features from medical imaging data to predict clinical variables. Various features can be constructed to describe the properties of the full image, or those of a specific region of interest such as a tumor. These features may be related to a wide variety of clinical variables, such as disease characteristics, genetics, and therapy response. This can be done through the use of machine learning, which enables the training of a model on these features using data of patients for which the relevant clinical variables are already known. The resulting models may be used as a diagnostic aid for the prediction of labels such as tumor phenotype and therapy response in new patients. Thereby, radiomics can provide a noninvasive alternative for invasive procedures, such as biopsies, to uncover disease characteristics or clinical outcomes. Radiomics therefore has a high potential to be a valuable tool for clinical practice. This may explain the rise in popularity of radiomics in the medical imaging research field in recent years, resulting in many methods and applications. This chapter provides a guide through the several aspects of designing a radiomics study.

References (0)

Cited by (36)

  • Modality-level cross-connection and attentional feature fusion based deep neural network for multi-modal brain tumor segmentation

    2023, Biomedical Signal Processing and Control
    Citation Excerpt :

    Manual brain tumor segmentation requires a lot of expertise, and the segmentation is often achieved in a slice-by-slice manner. Although we can take advantage of the expert knowledge, this method is very tedious and time-consuming, and the results are very subjective [4,5]. In the last few years, a large number of automated approaches have been proposed to segment the brain tumors, which can be generally categorized into two groups:

  • Deep Network Design for Medical Image Computing: Principles and Applications

    2023, Deep Network Design for Medical Image Computing: Principles and Applications
  • OpenFiberSeg: Open-source segmentation of individual fibers and porosity in tomographic scans of additively manufactured short fiber reinforced composites

    2022, Composites Science and Technology
    Citation Excerpt :

    As imaging apparatus are limited to a voxel size of 0.7 to 0.4 μm, fiber identification must be performed on at best a handful of voxels, as they typically have a diameter ranging from 5 to 10 μm. Manual labelling by an expert is possible, but is a tedious, time-consuming task, and subject to inter and intra-observer variability [37]. Automatic segmentation tools are therefore required.

  • Toward High-Throughput Artificial Intelligence-Based Segmentation in Oncological PET Imaging

    2021, PET Clinics
    Citation Excerpt :

    Accurate segmentation is also crucial for external beam therapy planning. In the last decade, the valuable role of radiomics4,5 for image assessment and outcome prediction has been reported, for which segmentation is a vital step.6–9 In clinical workflows, in the context of radiopharmaceutical therapies, segmentation of PET and/or single-photon emission computed tomographic (SPECT) images is also needed for image-based dosimetry as well as quantification of therapy response based on pretherapeutic and posttherapeutic images (see Julia Brosch-Lenz and colleagues’ article, “Role of AI in Theranostics: Towards Routine Personalized Radiopharmaceutical Therapies,” in this issue).

  • Role of Artificial Intelligence in Theranostics:: Toward Routine Personalized Radiopharmaceutical Therapies

    2021, PET Clinics
    Citation Excerpt :

    Segmentation of all these lesions manually is not practical. Manual segmentation is also subject to intraobserver53 and interobserver variability.54 Validated AI-based models for fully automated, robust, accurate segmentation of organs/lesions in PET, PET/CT, and SPECT/CT images can help delineate OARs and lesions to achieve a personalized dosimetry framework.

View all citing articles on Scopus
c

The first two authors are shared first authors.

View full text