Virtual Library
Start Your Search
W. Crijns
Author of
-
+
P1.03 - Poster Session/ Treatment of Locoregional Disease – NSCLC (ID 212)
- Event: WCLC 2015
- Type: Poster
- Track: Treatment of Locoregional Disease – NSCLC
- Presentations: 1
- Moderators:
- Coordinates: 9/07/2015, 09:30 - 17:00, Exhibit Hall (Hall B+C)
-
+
P1.03-021 - Lung Damage Quantification on CT Scans Strengthens Radiation-Induced Lung Toxicity Prediction Models (ID 2932)
09:30 - 17:00 | Author(s): W. Crijns
- Abstract
Background:
Predictive models for radiation-induced lung toxicity have shown a lack of validation and low values of area under the curve (AUC) below 0.7, for various reasons. Radiation-induced lung tissue damage scored as density changes on CT scans proved to be a less multifactorial endpoint compared to dyspnea. Its continuous variation in the patient population is an indication that it could be an expression of patient-specific radiosensitivity variation. This study explores the advantage of incorporating patient-specific lung damage measures in the classical predictive models based on mean lung dose (MLD).
Methods:
61 stage I-IV lung cancer patients treated with chemoradiotherapy were retrieved from two hospitals. Prescribed dose was 66 Gy in fractions of 2 Gy (concurrent) or 2.75 Gy (sequential). Baseline and follow-up dyspnea scores were retrospectively assessed according to CTCAE 4.0. Image analysis of the radiation-induced lung damage was performed by comparison of the baseline planning CT~0~ and the non-rigidly registered follow-up CT~fup~. The median Hounsfield Unit increase (∆HU=HU~fup~-HU~0~) was calculated per dose bin of 5 Gy. The local dose-∆HU response curve was described using a sigmoidal model. This resulted in a sigmoidal parameter D~50~ (corresponding to 50% of the saturation level of ∆HU) for each patient, as an expression of the patient-specific lung tissue radiosensitivity. Logistic models predicting dyspnea increase with respect to the baseline score were then built using MLD and D~50~ as covariates. The likelihood-ratio identified significant differences between nested models.
Results:
Dyspnea score increase by 2 grades was observed in 9 patients (14,8%), while an increase by 1 grade was observed in 29 patients (47.5%). The average timepoint of CT~fup~ was 2.3 months after end of radiotherapy. For 51 patients the sigmoidal dose-∆HU fits were acceptable (sum of squared residuals below 10 HU per datapoint on average). 10 of these patients did not show any dose response in the analysed dose range. The 41 reacting patients showed large variation in D~50 ~(median: 34.8 Gy, range: 12.1 Gy-70.0 Gy) and were further analysed. Predictive models based on MLD alone had AUCs of 0.71 and 0.65 for dyspnea increase by 1 and 2 grades respectively. Incorporating the CT damage measure D~50~ as second covariate resulted in models with 0.73 and 0.83 respectively. The advantage of incorporating D~50~ was significant in the second fit (p=0.05).
Conclusion:
A significant improvement of predictive models for radiation-induced lung toxicity was achieved using patient-specific lung damage measured on CT scans. An early detection of the patient-specific D~50~ through dedicated per-treatment imaging optimized for the detection of lung tissue changes is crucial for the clinical implementation of the model. Future work analysing more CT features could also improve the model.
