DL-Dose Framework: Introducing the Fast Dose Calculation Engine for Radiotherapy-Based Deep Learning

Articles in Press

Document Type : Original Paper

Authors

1 Department of Physics and Energy Engineering, Amir Kabir University of Technology, Iran

2 Radiation Oncology Research Centre, Cancer Institute, Tehran University of Medical Sciences, Tehran, Iran

10.22038/ijmp.2025.84901.2493

Abstract

Introduction

The objective of this survey is to develop a deep-learning dose calculation framework that utilizes water dose distribution and the characteristics of heterogeneous phantoms as inputs.

Methods

Two hundred heterogeneous phantoms with identical geometry but varying thicknesses of bone and lung were simulated using the Dosxyznrc code. The thicknesses of the bone and lung inhomogeneity layers were randomly varied, ranging from 1 cm to 5 cm at different locations along the Z-axis of the heterogeneous phantom. Then, we developed the dose prediction framework based on the deep learning method, with five inputs and a single output channel, to predict dose distribution. Five input channels were included: a matrix representing water dose distribution, a matrix of mass density, a matrix of CT numbers, a matrix indicating the distance of each voxel from the central point of the radiation field, and a matrix containing values of zero or one that corresponds to the radiation field. The output consisted of the dose distribution for each heterogeneous phantom.

Results

The accuracy of predicted results by the DL-Dose Framework was compared with those obtained by the Monte Carlo method, utilizing the delta index, inspired by the global gamma index in the heterogeneous phantoms. The dose distribution results in the water medium, before the heterogeneities, indicated that 100% of the dose distribution of voxels with deviations of less than 1% from the maximum dose were comparable to and the same as the results measured using the Monte Carlo method. Additionally, 94.2% of the dose distribution for voxels in areas of lung heterogeneity and 98.1% for voxels in regions of bone heterogeneity are comparable to the results obtained using Monte Carlo simulations, with deviations of less than 1% from the maximum dose.

conclusions

The developed DL-Dose Framework can predict dose distribution in heterogeneous phantoms accurately.

Keywords

Main Subjects

Iranian Journal of Medical Physics

Articles in Press, Accepted Manuscript
Available Online from 05 October 2025

Files

History

  • Receive Date: 02 January 2025
  • Revise Date: 02 August 2025
  • Accept Date: 05 October 2025

Share

How to cite

Statistics