University of Twente Student Theses


Automatic Lung Aeration Assessment for Pediatric Lung Ultrasound Imaging

Logendran, T. (2022) Automatic Lung Aeration Assessment for Pediatric Lung Ultrasound Imaging.

[img] PDF
Abstract:Background: Imaging of the lungs and chest is traditionally done with chest X-ray (CXR) imaging as a first-line imaging method, but due to the irradiation exposure this is harmful for patients in the pediatric intensive care unit (PICU). Lung ultrasonography (LUS) is a suitable alternative, but it requires much practice and sufficient skill to interpret the images. Therefore the aim of this study was to develop deep learning software to assist the interpreter in LUS assessments. Methods: 506 LUS videos were collected from 33 PICU patients. A recurrent convolutional neural network (RCNN) was developed to assign input videos a LUS score. A generative adversarial network (GAN) was developed to generate segmentation masks containing clinical features from individual LUS frames. Results: For the classification tasks the average sensitivity was 0.49, specificity 0.84, accuracy 0.79, precision 0.51 and F1-score 0.47. Performance results of the GAN were an average dice similarity coefficient (DSC) of 0.97 ± 0.029 and a mean squared error (MSE) of 0.025 ± 0.025. Conclusions: Poor sensitivities make that the RCNN is not ready to be used in clinical practice yet, the GAN is moderately able to localize important features. When both models have achieved performances that are sufficient for clinical implementation, LUS can more easily be used by clinical professionals.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general, 54 computer science
Programme:Technical Medicine MSc (60033)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page