Quantitative Analysis and Automated Lung Ultrasound Scoring for Evaluating COVID-19 Pneumonia With Neural Networks

Abstract

As being radiation-free, portable, and capable of repetitive use, ultrasonography is playing an important role in diagnosing and evaluating the COVID-19 Pneumonia (PN) in this epidemic. By virtue of lung ultrasound scores (LUSS), lung ultrasound (LUS) was used to estimate the excessive lung fluid that is an important clinical manifestation of COVID-19 PN, with high sensitivity and specificity. However, as a qualitative method, LUSS suffered from large interobserver variations and requirement for experienced clinicians. Considering this limitation, we developed a quantitative and automatic lung ultrasound scoring system for evaluating the COVID-19 PN. A total of 1527 ultrasound images prospectively collected from 31 COVID-19 PN patients with different clinical conditions were evaluated and scored with LUSS by experienced clinicians. All images were processed via a series of computeraided analysis, including curve-to-linear conversion, pleural line detection, region-of-interest (ROI) selection, and feature extraction. A collection of 28 features extracted from the ROI was specifically defined for mimicking the LUSS. Multilayer fully connected neural networks, support vector machines, and decision trees were developed for scoring LUS images using the fivefold cross validation. The model with 128 × 256 two fully connected layers gave the best accuracy of 87%. It is concluded that the proposed method could Manuscript

Automated lung ultrasound scoring for evaluation of coronavirus disease 2019 pneumonia using two-stage cascaded deep learning model

“…Previously, Carrer et al [9] proposed a pleural line detection and localization method which employed the eight features of pleural line and its underlying area into the machine learning for the automated scoring. Chen et al [29] also designed a similar scoring model including five steps, Step 1 transferred image format, Step 2 performed the pleural line detection, Step 3 selected the ROI, Step 4 extracted 28 different features, and Step 5 achieved the automatic scoring (Scores 0–3) based on deep learning. This model analyzed more features than Carrer’s method [9] and obtained a great performance for scoring LUS images.…”

“…For example, Carrer et al [9] employed the eight features of pleural line and its underlying area into the machine learning for the automatic LUS scoring. Chen et al [29] proposed a quantitative feature extraction method, and used the neural networks, support vector machines, and decision trees for automatic scoring of LUS images. Wang et al [30] designed four features related to pleural line and four features related to B-line to analyze the lung ultrasound images, and realized the binary severe/non-severe classification by support vector machine.…”

Quantitative Analysis of Pleural Line and B-Lines in Lung Ultrasound Images for Severity Assessment of COVID-19 Pneumonia

“…But these studies mainly focused on the intensity-related properties of the pleural line or the distribution of B-lines, and did not consider the pleural line-and B-line-related characteristics simultaneously. Some deep learning based techniques have shown promises in automatic detection of B lines and pleural effusion, binary classification and 4-level scoring system using artificially extracted quantitative parameters, frame-level and video-level ultrasound images or combination of images and clinical information as the input [33,34,36,46,47], but they need a large number of annotated samples and have poor interpretability. As a contrast, the quantitative method we propose aims at characterizing the image patterns of clinical findings and digging out more features from the pleural line and B-lines, which are further combined to provide more comprehensive information for the severity assessment.…”

“…The delineated region should be large enough to include the entire pleural line or B-line visible in the field of view, which is helpful to ensure the consistency between different patients. Many automatic methods for detection of the pleural line and localization of B-lines have been proposed based on various image processing algorithms and deep learning networks [20,32,33,35,48,[51][52][53][54][55]. In the future, the pleural-line and B-lines will be full-automatically detected to develop a more intelligent diagnostic system for COVID-19 pneumonia.…”

“…the scoring systems based on visual observations of the line artifacts and quantitative methods based on automatic detection of the pleural line and B-lines have been proposed for the diagnosis, evaluation, and follow-up of the patients of COVID-19 pneumonia. Nevertheless, these approaches mainly focus on the features related to the characteristics of the pleural line and Blines separately [20,32,33]. Characteristics associated with the pleural line and B-lines are not combined to evaluate the lung conditions.…”