Click here to

Session: Data Science, Radiomics, and Computing [Return to Session]

A Radiomics-Boosted Deep Learning Model for COVID-19 and Non-COVID-19 Pneumonia Detection Using Chest X-Ray Image

Z Hu1*, Z Yang2, F Yin2, K Lafata2, C Wang2, (1) Duke Kunshan University, Kunshan, Jiangsu, China, (2) Duke University Medical Center, Durham, NC


SU-E-TRACK 6-6 (Sunday, 7/25/2021) 3:30 PM - 4:30 PM [Eastern Time (GMT-4)]

Purpose: To develop a deep-learning model that integrates radiomics analysis for enhanced performance of COVID-19 and Non-COVID-19 pneumonia detection using chest X-ray image

Methods: Based on a pre-trained VGG-16 architecture, the deep-learning design consists of a 5-Dense layer neural network with varying sizes, while the last three convolutional layers were set as free parameters for training. In radiomics analysis, a 2D sliding kernel was implemented to map the impulse response of radiomic features throughout the entire X-ray image; thus, each feature is rendered as a 2D map in the same dimension as X-ray image. Two deep-learning models were trained: in the 1st model, X-ray image was the sole input; in the 2nd model, X-ray image and 2 radiomic feature maps(RFM) selected by the saliency map analysis of the 1st model were stacked as the input. Both models were developed using 812 chest X-ray images with 262/288/262 COVID-19/Non-COVID-19 pneumonia/healthy cases, and 649/163 cases were assigned as training-validation/independent test sets. For each model design, 25 versions were trained with random assignments of training/validation cases following 7:1 ratio in training-validation set. Sensitivity, specificity, accuracy, and ROC curve results from both model designs were analyzed.

Results: In 1st model using X-ray as the sole input, the 1)sensitivity, 2)specificity, 3)accuracy, and 4)ROC Area-Under-the-Curve (AUC) of COVID-19/Non-COVID-19 pneumonia detection were 1)0.91±0.08/0.78±0.07, 2) 0.93±0.04/0.94±0.05, 3)0.92±0.03/0.89±0.02, and 4)0.96±0.03/0.92±0.03. In the 2nd model design, two RFMs, Entropy and Short-Run-Emphasize, were selected with their highest cross-correlations with the saliency maps of the 1st model. The corresponding results demonstrated significant improvements(p<0.05) of disease detection: 1)0.95±0.03/0.85±0.04, 2)0.97±0.02/0.96±0.02, 3)0.97±0.02/0.93±0.02, and 4)0.99±0.01/0.97±0.02. The reduced variations suggested a superior robustness of 2nd model design.

Conclusion: The inclusion of radiomic analysis in deep-learning design improved the performance and robustness of COVID-19/Non-COVID-19 pneumonia detection, which holds great potential for clinical applications in the COVID-19 pandemic.



    CAD, Modeling, Image Analysis


    IM/TH- Image Analysis (Single Modality or Multi-Modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

    Contact Email