Abstract
Radiation pneumonitis (RP) is one of the major toxicities of thoracic radiation therapy. RP incidence has been proven to be closely associated with the dosimetric factors and normal tissue control possibility (NTCP) factors. However, because these factors only utilize limited information of the dose distribution, the prediction abilities of these factors are modest. We adopted the dosiomics method for RP prediction. The dosiomics method first extracts spatial features of the dose distribution within ipsilateral, contralateral, and total lungs, and then uses these extracted features to construct prediction model via univariate and multivariate logistic regression (LR). The dosiomics method is validated using 70 non-small cell lung cancer (NSCLC) patients treated with volumetric modulated arc therapy (VMAT) radiotherapy. Dosimetric and NTCP factors based prediction models are also constructed to compare with the dosiomics features based prediction model. For the dosimetric, NTCP and dosiomics factors/features, the most significant single factors/features are the mean dose, parallel/serial (PS) NTCP and gray level co-occurrence matrix (GLCM) contrast of ipsilateral lung, respectively. And the area under curve (AUC) of univariate LR is 0.665, 0.710 and 0.709, respectively. The second significant factors are V5 of contralateral lung, equivalent uniform dose (EUD) derived from PS NTCP of contralateral lung and the low gray level run emphasis of gray level run length matrix (GLRLM) of total lungs. The AUC of multivariate LR is improved to 0.676, 0.744, and 0.782, respectively. The results demonstrate that the univariate LR of dosiomics features has approximate predictive ability with NTCP factors, and the multivariate LR outperforms both the dosimetric and NTCP factors. In conclusion, the spatial features of dose distribution extracted by the dosiomics method effectively improves the prediction ability.
Introduction
In thoracic radiation therapy, Radiation Pneumonitis (RP) remains a significant clinical challenge. Current predictive standards rely heavily on dosimetric factors (e.g., Mean Lung Dose) and Normal Tissue Complication Probability (NTCP) models. These models are primarily derived from the Dose-Volume Histogram (DVH), which reduces complex 3D dose distributions into a 1D summary. Consequently, they fail to account for the spatial heterogeneity of the dose—two patients with identical DVHs might have vastly different spatial dose patterns, leading to different clinical outcomes. This study introduces "dosiomics," a framework that applies texture analysis to 3D dose grids to capture these critical spatial features.
Methods
The study analyzed 70 non-small cell lung cancer (NSCLC) patients treated with Volumetric Modulated Arc Therapy (VMAT). The clinical endpoint was Grade ≥ 2 radiation pneumonitis.
Feature Extraction and Modeling
A total of 129 spatial features were extracted from the dose distributions of the ipsilateral, contralateral, and total lungs. These features included:
- Gray Level Co-occurrence Matrix (GLCM): Captures local dose variations between adjacent voxels.
- Gray Level Run Length Matrix (GLRLM): Identifies regional patterns, such as sequences of similar dose levels.
Prediction models were constructed using univariate and multivariate Logistic Regression (LR). To prevent overfitting, a bootstrap method (1,000 iterations) was employed, and redundant features were excluded via Spearman correlation analysis. The dosiomics models were compared against standard dosimetric factors ($V_5, V_{20}$, Mean Lung Dose) and established NTCP models (Lyman and Parallel/Serial models).
Results
The dosiomics approach demonstrated superior predictive performance compared to traditional methods.
Comparative Performance
- Univariate Analysis: The best single dosiomics feature (GLCM contrast of the ipsilateral lung) achieved an AUC of 0.709, comparable to the best NTCP factor (AUC 0.710) and superior to the best dosimetric factor (Mean Dose, AUC 0.665).
- Multivariate Analysis: The multivariate dosiomics model significantly outperformed its counterparts:
- Dosiomics Model: AUC = 0.782
- NTCP Model: AUC = 0.744
- Dosimetric Model: AUC = 0.676
Key Predictive Features
The most significant single feature was "GLCM contrast" in the ipsilateral lung, where a one-unit increase was associated with a 2.01-fold increase in RP risk. Additionally, "low gray level run emphasis" in the total lungs—representing large areas of low dose—emerged as a critical secondary predictor in the multivariate model.
Discussion
The findings highlight that spatial heterogeneity and the extent of low-dose regions are vital indicators of lung injury. Unlike traditional factors which often provide redundant information, dosiomics features offer a diverse set of descriptors that enhance multivariate predictive power.
While the study is limited by a relatively small sample size (n=70), the use of bootstrapping provides a robust initial validation. Future research should explore "pan-omics" integration—combining dosiomics with radiomics and genomics—and investigate the potential for using these spatial features to guide treatment planning to actively minimize toxicity risks.
Conclusion
Dosiomics effectively captures the 3D spatial characteristics of dose distributions that are missed by traditional DVH-based models. By incorporating these features, clinicians can significantly improve the prediction of radiation pneumonitis, paving the way for more personalized and safer thoracic radiotherapy.
Summary updated