An interpretable deep learning workflow for discovering subvisual abnormalities in CT scans of COVID-19 inpatients and survivors

Abstract

Tremendous efforts have been made to improve diagnosis and treatment of COVID-19, but knowledge on long-term complications is limited. In particular, a large portion of survivors has respiratory complications, but currently, experienced radiologists and state-of-the-art artificial intelligence systems are not able to detect many abnormalities from follow-up computerized tomography (CT) scans of COVID-19 survivors. Here we propose Deep-LungParenchyma-Enhancing (DLPE), a computer-aided detection (CAD) method for detecting and quantifying pulmonary parenchyma lesions on chest CT. Through proposing a number of deep-learning-based segmentation models and assembling them in an interpretable manner, DLPE removes irrelevant tissues from the perspective of pulmonary parenchyma, and calculates the scan-level optimal window, which considerably enhances parenchyma lesions relative to the lung window. Aided by DLPE, radiologists discovered novel and interpretable lesions from COVID-19 inpatients and survivors, which were previously invisible under the lung window. Based on DLPE, we removed the scan-level bias of CT scans, and then extracted precise radiomics from such novel lesions. We further demonstrated that these radiomics have strong predictive power for key COVID-19 clinical metrics on an inpatient cohort of 1,193 CT scans and for sequelae on a survivor cohort of 219 CT scans. Our work sheds light on the development of interpretable medical artificial intelligence and showcases how artificial intelligence can discover medical findings that are beyond sight.

DLKN-MLC: A Disease Prediction Model via Multi-Label Learning

“…At the same time, most of the current studies are end-to-end models [ 43 ]; compared with such studies, the DLKN-MLC model has better interpretability [ 44 ]. We use the DLKN-MLC model to simulate the process of doctors obtaining patient information and making diagnosis inferences, and we can obtain the information of DLKN-MLC used to infer disease through the matching information of patient disease information extracted from the EHR and binary-weighted network.…”

Six-month follow-up after recovery of COVID-19 Delta variant survivors via CT-based deep learning

“…Namely, the workflow of this model only allows marking visual lesions on regular CT scans, but not sub-visual lesions (i.e., it is almost impossible for a radiologist to see fibrosis lesions directly from ordinary CT scans). Among our team's latest published techniques ( 7 ), the deep lung parenchyma enhancing (DLPE) method was used to automatically mark visible and sub-visual COVID-19 lesions. In the follow-up study, we used the DLPE method to avoid the lesion-omissions issue that might occur in similar studies with traditional AI applications.studies.…”

“…We collected these patients' last CT scan and clinical data before they were discharged from the hospital and continued to collect CT scans and important clinical indicators during the 6-month follow-up. Further, we used AI approaches such as deep lung parenchyma enhancing (DLPE) to quantify follow-up CT and discharged CT ( 7 ) and to provide a comprehensive assessment of patient recovery and prognosis. Our findings provide intrinsic insights into the mechanisms underlying the prognosis of COVID-19, especially the Delta variant.…”

Segmentation of Lung Lobes and Lesions in Chest CT for the Classification of COVID-19 Severity

“…Subsequently, a U-Net model was created to support three-dimensional (3D) matrices and is called 3D-Unet 23 . The 3D-UNet model was used to develop a more e cient 3D imaging model for the segmentation of lesions and lung tissue 24,25 . Cropping the lung area before lesion segmentation can improve accuracy 26 .…”