Identification of Robust Normal Lung CT Texture Features
Download as PPTX, PDF0 likes3,023 views
The document discusses the identification of 11 robust normal lung CT texture features that can predict radiation-induced lung disease during radiotherapy, specifically radiation pneumonitis and fibrosis. These features are relatively invariant to tumor size and do not correlate with normal lung volume, highlighting their clinical utility. The study involved analyzing texture features from simulated normal lung volumes without gross tumor volume (GTV) and was partially supported by NIH funding.
Identification of Robust Normal Lung CT Texture Features
- 1. Identification of Robust Normal Lung CT Texture Features for the Prediction of Radiation-Induced Lung Disease Wookjin Choi, Sadegh Riyahi, Wei Lu Medical Physics, Memorial Sloan Kettering Cancer Center Radiation Oncology, University of Maryland School of Medicine Purpose • Radiomics features for normal tissue in lung CT can help predict radiation induced lung disease during radiotherapy – Radiation pneumonitis and Radiation fibrosis • Clinically useful features – Relatively invariant (robust) to tumor size as well as not correlated with normal lung volume Results Conclusion Supported in part by NIH R01CA172638. *Contact: Wei Lu, Ph.D., luw@mskcc.org (a) (b) Figure 3. (a) The feature variations with respect to tumor size (the diameter of the GTV) in robust and unrobust features; (b) A small normal lung in axial CT that caused by atelectasis (red arrow); right lung – cancerous, left lung – normal, and green circular hole – 60-mm diameter GTV. Figure 2. (a) Distributions of feature variations for each feature, the red line (5%) is the robustness threshold; (b) Correlations between each texture feature and the volume of the simulated normal lung without GTV • We identified 11 robust normal lung CT texture features. • The robust features can be further examined for the prediction of radiation-induced lung disease. • Interestingly, low grey-level run features identified normal lung diseases. • Only 11 features were robust. – All first-order intensity-histogram features (min, max, mean, and median), two of the GLCM and four of the GLRM features were robust. • Correlation with normal lung volume – All robust features were not correlated – Three unrobust features showed high correlation • Larger GTV resulted greater feature variation • There was no dependence on GTV location • Excessive variations were observed – Two low grey-level run features – Identified local lung diseases (atelectasis)Figure 1. A flow chart for the normal lung CT texture features’ robustness analysis Different sizes of GTVs were simulated at the normal lung (contralateral to tumor). Simulation for Normal lung without GTV 20, 30, 40, 50 and 60 mm diameter sphere at upper or lower lobe GTV(hole) Generation Reference Lung CT features Reference Lung CT features Simulated Lung CT features Robustness Analysis Results Lung CT Lung Contour Texture Feature Extraction 𝑭𝒓𝒆𝒇 𝑭𝒔𝒊𝒎 𝑫 = 𝑭 𝒓𝒆𝒇 − 𝑭 𝒔𝒊𝒎 𝑭 𝒓𝒆𝒇 ∙ 𝟏𝟎𝟎% • Feature robustness evaluation – Relative difference – Pearson correlation Method • The free-breathing CTs of 14 lung SBRT patients were studied. • 27 texture features were extracted from simulated normal lung volume – – 9 intensity histogram based features – 8 grey-level co-occurrence matrix (GLCM) features – 10 grey-level run-length matrix (GLRM) features 𝑉𝑠𝑖𝑚 = 𝑉𝑙𝑢𝑛𝑔 − 𝑉𝐺𝑇𝑉