Predicting COVID-19 Lung Infiltrate Progression on Chest Radiographs Using Spatio-temporal LSTM based Encoder-Decoder NetworkDownload PDF

Feb 10, 2021 (edited Feb 22, 2021)MIDL 2021 Conference SubmissionReaders: Everyone
  • Abstract: Automated analyses of chest imaging in Coronavirus Disease 2019 (COVID-19) have largely focused on a single timepoint, usually at disease presentation, and have not explicitly taken into account temporal disease manifestations. We present a deep learning-based approach for prediction of imaging progression from serial chest radiographs (CXRs) of COVID-19 patients. Our method first utilizes convolutional neural networks (CNNs) for feature extraction from patches within the concerned lung zone, and also from neighboring areas to enhance the contextual phenotypic information. The framework further incorporates two distinct spatio-temporal Long Short Term Memory (LSTM) modules for effective predictions. The first LSTM module captures spatial dependencies between patches and the second exploits the temporal context of sequential CXR scans. The resulting network focuses on critical image regions that provide relevant information for learning the progression of lung infiltrates without the explicit need for infiltrate segmentation. The second LSTM provides an encoded context vector used as an input to a decoder module to predict future severity grades. Our novel multi-institutional dataset comprises sequential CXR scans from N=100 patients. Specifically, our framework predicts zone-wise disease severity for a patient on the last day by learning representations from the previous temporal CXRs. We design two baseline approaches - one using fine-tuned VGG-16 features and the other using radiomic descriptors. Experimental results demonstrate that our proposed approach outperforms both baselines in average accuracy by 10.33% and 12.16%, respectively, in predicting COVID-19 progression severity.
  • Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
  • Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
  • Paper Type: both
  • Source Latex: zip
  • Primary Subject Area: Application: Radiology
  • Secondary Subject Area: Validation Study
15 Replies

Loading