# Multispectral PCCT and CBCT imaging for high precision radiotherapy through translation of imaging parameters with machine learning validation

## TL;DR

This study explores how multispectral CT imaging can be used in high-precision radiotherapy by comparing different imaging techniques and validating results with machine learning.

## Contribution

The study introduces a novel method to translate PCCT parameters to CBCT using machine learning, enabling better integration in radiotherapy.

## Key findings

- CBCT (iCBCT Acuros) showed stronger agreement with PCCT-derived 60 and 67 keV VMI than T3D.
- Machine learning confirmed alignment between CBCT and PCCT-based VMI, but not T3D.
- Image quality was highest for T3D PCCT, and CCC values were significantly affected by CBCT presets.

## Abstract

Photon-counting CT (PCCT) is the mainstay of multi-spectral imaging, enabling quantitative tissue characterization. In radiation oncology, cone-beam CT is used daily for image-guided and online-adaptive radiotherapy. The novel HyperSight cone-beam CT imaging mode (CBCT), with enhanced image quality due in part to its enlarged detector size and optimized reconstruction modes, further facilitates quantitative image monitoring and high-precision radiotherapy. Integrating spectral PCCT information may further amplify its potential. Therefore, this study investigates whether qualitative and spectral quantitative PCCT-parameters can be translated to CBCT. An inorganic tissue-equivalent anthropomorphic phantom analysis was conducted using CBCT (iCBCT/iCBCT Acuros reconstruction, Pelvis/Pelvis Large preset) and PCCT (T3D (polychromatic reconstruction) with virtual monochromatic imaging (VMI)). Twenty regions with different CT numbers were assessed qualitatively and quantitatively. Image quality was highest for T3D PCCT. Quantitative analysis showed stronger agreement between CBCT (iCBCT Acuros) and PCCT-derived 60 and 67 keV VMI (concordance correlation coefficient (CCC) ≥ 0.595), compared to T3D (CCC ≤ 0.183), with CCC values significantly affected by CBCT presets and reconstruction method (p ≤ 0.001). Machine learning-based hierarchical clustering confirmed alignment between CBCT and PCCT-based VMI, but not T3D. This successful translatability of specific VMI levels paves the way for the integration of multi-spectral imaging into high-precision CBCT-based radiotherapy using PCCT.