# Classification of anatomic structures in head and neck by CT-based radiomics

**Authors:** Yoichi Watanabe, A. Biswas, K. Rangarajan, G. Rath, N. Gopishankar  
**arXiv:** [2205.08054](https://arxiv.org/abs/2205.08054)  
**Date:** 2022-05-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2205.08054v1/

## Key points

- Investigated differences in radiomics features among various anatomical structures in the head and neck area using CT images
- Explored whether CT radiomics could be used to classify anatomical structures using unsupervised machine-learning techniques
- Analyzed treatment planning data from 36 patients with head and neck cancers, totaling 1357 contours of more than 22 anatomical structures
- Calculated 174 radiomics features for each structure using the SIBEX program
- Tested uniqueness of radiomics features for classifying contours into their respective anatomical groups
- Developed a two-stage clustering technique to further classify the 22 anatomical structures into sub-groups with similar characteristics
- Identified distinct differences in heatmap of radiomics features between tumors and healthy structures in head and neck area
- Achieved over 90% accuracy in identifying specific structures such as eyes, lens, submandibular glands, pituitary glands, and thyroids using radiomics features
- Identified six subgroups among the 22 anatomical structures based on shared characteristics such as fatty and bony tissues through clustering technique
- Demonstrated potential of utilizing radiomics features for disease identification and therapy outcome prediction in clinical practice for head and neck tumors

## Summary

In this study, the authors aimed to investigate the differences in radiomics features among various anatomical structures in the head and neck area using CT images. They also explored whether CT radiomics could be used to classify these anatomical structures using unsupervised machine-learning techniques. The researchers obtained treatment planning data from 36 patients who were treated for head and neck cancers at a single institution. They analyzed a total of 1357 contours of more than 22 anatomical structures drawn on planning CT scans. Using the SIBEX program, they calculated 174 radiomics features for each structure. First, they tested whether these radiomics features were unique enough to classify all contours into their respective anatomical groups. Then, they developed a two-stage clustering technique to further classify the 22 anatomical structures into sub-groups with similar physiological or biological characteristics. The results showed that there was a distinct difference in the heatmap of the 174 radiomics features among tumors and other healthy structures in the head and neck area. The researchers were able to identify specific structures such as the eyes, lens, submandibular glands, pituitary glands, and thyroids with over 90% accuracy using these radiomics features. Furthermore, through the two-stage clustering technique, they identified six subgroups among the 22 anatomical structures based on shared characteristics such as fatty and bony tissues. Overall, this study demonstrated that different anatomical structures in head and neck tumors have distinguishable radiomics features which can be utilized for disease identification and therapy outcome prediction in clinical practice. The findings suggest that CT radiomics can provide valuable insights into the biological characteristics of head and neck lesions; thus contributing to our understanding of how radiomics can be utilized for clinical purposes.