Prediction of Radiation Pneumonitis for NSCLC Using Crop Convolutional Neural Network Based Model
D Kawahara1*, N Imano1, R Nishioka2, A Saito1, Y Nagata1, (1) Hiroshima University,Hiroshima, JP, (2) Medical and Dental Sciences Course, Graduate School of Biomedical & Health Sciences, Hiroshima University, Hiroshima, JP
Presentations
WE-IePD-TRACK 6-2 (Wednesday, 7/28/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]
Purpose: To predict grade 2 radiation pneumonitis (RP) in locally advanced non-small cell lung cancer (NSCLC) patients with high accuracy by using convolutional-neural network (CNN) model.
Methods: Three CNN models were created: the whole body region, normal lung (nLung) region, and nLung region overlapping the region over 20 Gy (nLung∩20Gy) were fed into 3D CNN model. Models were evaluated on 54 patients and were independently tested on 23 newly treated patients. The risk of RP was categorized into two groups (Group I: more than RP grade 2 and Group II: whithin RP grade1). The model was optimized using Adam. Fivefold cross-validation was applied to verify its predictive performance. The precision, accuracy, and sensitivity by generating confusion matrices and the areas under the receiver operating characteristic curve (AUC) for each model were evaluated.
Results: The accuracy, specificity, sensitivity, and AUC were 53.9%, 80.0%, 25.5%, and 0.58, respectively for whole body method and 60.0%, 81.7%, 36.4%, and 0.64, respectively for nLung method. For the nLung∩20Gy method, the accuracy, specificity, sensitivity, and AUC were improved to 75.7%, 80.0%, 70.9%, and 0.84, respectively. The segmented input data can be learned effectively and useful for CNN model.
Conclusion: The CNN model which the input image was segmented and considered dose distribution can help to predict grade 2 RP for NSCLC after definitive radiotherapy.
ePosters
Keywords
Imaging Theory, CT, Radiation Therapy
Taxonomy
Not Applicable / None Entered.
Contact Email