Classification of common thoracic disorders from X-ray images

Bc. Enkhtaivan Ganbat

Master's thesis

Abstract:

I have trained and tested two models based on the ResNet-50 and MobileNet models. The models have been trained on the ChestX-ray14 dataset, which is currently one of the largest publicly available dataset of X-ray images, containing over 112 thousand images collected from the medical reports of nearly 31 thousand patients. Each X-ray image is labelled either ‘No Findings’ (normal) or with at least one of the 14 thoracic disorders. AUC (Area Under Curve) scores were used to measure model accuracy. Furthermore, three Voting Classifiers: Max Vote, Hard Vote, Simple Average were used along with AUC scores from the two models in an attempt achieve higher classification accuracy.

Keywords

Thoracic disorder, X-ray images, classification, multi-label, convolutional neural network, machine learning, deep learning, computer vision

Language used: English

Date on which the thesis was submitted / produced: 29. 3. 2021

Thesis defence

• Supervisor: doc. Ing. Arnošt Veselý, CSc.

Czech University of Life Sciences Prague

Faculty of Economics and Management

Master programme / field:
Systems Engineering and Informatics / Informatics