Potential diagnosis of COVID-19 from chest X-ray and CT findings using semi-supervised learning

Objective: Literature related to the imaging of COVID-19 pneumonia, its findings and contribution to diagnosis and its differences from adults are limited in pediatric patients. The aim of this study was to evaluate chest X-ray and chest CT findings in children with COVID-19 pneumonia. Methods: Chest X-ray findings of 59 pediatric patients and chest CT findings of 22 patients with a confirmed diagnosis of COVID-19 pneumonia were evaluated retrospectively. Results: COVID-19 pneumonia was most commonly observed unilaterally and in lower zones of lungs in chest X-ray examinations. Bilateral and multifocal involvement (55%) was the most observed involvement in the CT examinations, as well as, single lesion and single lobe (27%) involvement were also detected. Pure ground-glass appearance was observed in 41%, ground-glass appearance and consolidation together was in 36%. While peripheral and central co-distribution of the lesions (55%) were frequently observed, the involvement of the lower lobes (69%) was significant. In four cases,the coexistence of multiple rounded multifocal ground-glass appearance and rounded consolidation were observed. Conclusion: COVID-19 pneumonia imaging findings may differ in the pediatric population from adults. In diagnosis, chest X-ray should be preferred, CT should be requested if there is a pathologic finding on radiography that merits further evaluation and if clinically indicated. Advances in knowledge: Radiological findings of COVID-19 observed in children may differ from adults. Chest X-ray should often be sufficient in children avoiding additional irradiation, chest CT needs only be done in cases of clinical necessity.

Download Full-text

On Teacher-Student Semi-Supervised Learning for Chest X-ray Image Classification

10.21528/cbic2021-80 ◽

2021 ◽

Author(s):

Roberto Augusto Philippi Martins ◽

Danilo Silva

Keyword(s):

Deep Learning ◽

Supervised Learning ◽

Learning Algorithm ◽

Unlabeled Data ◽

Imaging Data ◽

X Ray ◽

Teacher Student ◽

Chest X Ray ◽

Deep Learning Model

The lack of labeled data is one of the main prohibiting issues on the development of deep learning models, as they rely on large labeled datasets in order to achieve high accuracy in complex tasks. Our objective is to evaluate the performance gain of having additional unlabeled data in the training of a deep learning model when working with medical imaging data. We present a semi-supervised learning algorithm that utilizes a teacher-student paradigm in order to leverage unlabeled data in the classification of chest X-ray images. Using our algorithm on the ChestX-ray14 dataset, we manage to achieve a substantial increase in performance when using small labeled datasets. With our method, a model achieves an AUROC of 0.822 with only 2% labeled data and 0.865 with 5% labeled data, while a fully supervised method achieves an AUROC of 0.807 with 5% labeled data and only 0.845 with 10%.

Download Full-text