Texture analysis using a self-organizing feature map

Alsyed, Emad, Smith, Rhodri, Marshall, Christopher and Spezi, Emiliano 2024. Texture analysis using a self-organizing feature map. El-Baz, Ayman, Ghazal, Mohammed and Suri, Jasjit S., eds. Handbook of Texture Analysis: AI-Based Medical Imaging Applications, Boca Raton: CRC Press, (10.1201/9780367486082-5)

Full text not available from this repository.

Abstract

The provision and clinical utility of texture analysis for quantification of intratumor heterogeneity in medical images is an established research topic. More recent attention has focused on the collective analysis of the several texture features available, in an omics-style approach of the radio-logical image. Radiomics analysis therefore encompasses methods of data extraction, feature(s) selection, and association analysis with pre-determined outcome metrics, such as tumor type, treatment response, and survival prognosis. Radiomics workflows are therefore rife with proposed artificial intelligence/machine learning solutions to extract and analyze as much meaningful quantitative data as possible, to be used in clinical decision support. Routine clinical imaging techniques and subsequent image analysis methodologies, however, demonstrate a wide variation in image acquisition/processing parameters and introduce numerous confounding variables into the radiomics pipeline. A lack of understanding on the dependency of texture features with these confounding variables serves as a roadblock to the urgent need to standardize texture analysis. A deeper understanding of the main confounding variables would allow the pooling of results from radiomic processed studies obtained from multi-center clinical trials. This would undoubtedly assist in the development of radiomics as a discipline and propel its success and acceptance clinically. To this avail, we introduce the application of a Kohonen self-organizing map (SOM) to radiomic analysis. The SOM serves as a powerful general-purpose exploratory instrument to reveal the statistical indicators of changes in texture distributions and allows the identification of emergent properties or behavior in texture features when performing image acquisition/processing. The SOM may also be used in combination with outcome metrics to serve as a radiomic predictive model to assist in clinical management. In this chapter, we examine the applicability of the SOM within the domain of positron emission tomography (PET) imaging.

Item Type:	Book Section
Date Type:	Publication
Status:	Published
Schools:	Engineering
Publisher:	CRC Press
ISBN:	9780367486082
Last Modified:	09 Sep 2024 14:00
URI:	https://orca.cardiff.ac.uk/id/eprint/171895

Actions (repository staff only)

Edit Item