Related Experiment Video
Updated: May 21, 2026
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Improving Diagnosis of Severe Paediatric Lymphobronchial Tuberculosis Using Segmentation and Deep Learning on Chest
Marthinus Basson1, Pierre Goussard2, André George Gie3
1Department of Industrial Engineering, Stellenbosch University, Cape Town, South Africa.
Insights
Diagnosing childhood tuberculosis (TB) in young children is difficult. A deep learning model using chest X-rays achieved high accuracy in identifying pulmonary TB in children under five.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Pediatric Infectious Diseases
Background:
- Diagnosing intrathoracic tuberculosis (TB) in children under five is challenging due to overlapping symptoms and difficulty in microbiological confirmation.
- Current diagnostic methods, including physician interpretation of chest X-rays (CXRs) showing lymphadenopathy, have moderate performance (sensitivity 67-74%, specificity 39-59%).
- There is a need for improved diagnostic tools specifically for this vulnerable pediatric population.
Purpose of the Study:
- To develop and evaluate an automated deep learning approach for identifying severe pulmonary lymphobronchial tuberculosis (LBTB) in children under five.
- To utilize anterior-posterior (AP) chest X-ray (CXR) images for diagnosis.
- To use confirmed airway compression as the primary radiographic reference standard.
Main Methods:
- A dataset of 402 AP chest X-ray images was curated.
- Images included children with confirmed intrathoracic TB and airway obstruction, and healthy children.
- Three convolutional neural network (CNN) architectures (VGG16, ResNet50, InceptionV3) were evaluated for LBTB classification.
Main Results:
- A fine-tuned ResNet50 model demonstrated high diagnostic performance.
- The model achieved a mean test sensitivity of 94.43% and specificity of 94.79%.
- The model successfully identified radiographic manifestations of LBTB affecting the trachea and main bronchi on CXRs.
Conclusions:
- Deep learning, specifically the ResNet50 model, shows significant promise for accurately diagnosing pulmonary TB in young children using CXR.
- This automated approach can aid clinicians in diagnosing challenging cases of pediatric intrathoracic TB.
- Further validation in larger, diverse cohorts is warranted to confirm generalizability.
Introduction:
Diagnosing intra-thoracic tuberculosis (TB) in children under 5 years of age remains challenging. Clinical symptoms often overlap with those of other common childhood illnesses, and microbiological confirmation is frequently difficult to obtain in this age group. Despite advances in diagnostic technologies, few studies focus specifically on children younger than 5 years, the population in whom diagnosis is most complex. Intra-thoracic lymphadenopathy on chest X-ray (CXR) is the hallmark radiological features of TB in this age group, but identification by trained physicians currently demonstrates only moderate diagnostic performance, with reported sensitivities ranging from 67% to 74% and specificities between 39% and 59%. We aimed to develop and evaluate an automated deep learning-based approach for identifying severe pulmonary lymphobronchial tuberculosis (LBTB) in children under 5 years of age using anterior-posterior (AP) CXR images, with confirmed airway compression serving as the primary radiographic reference.
Methods:
A total of 402 AP CXR images were included in this study. Half of the images were obtained from children diagnosed with intra-thoracic TB and with confirmed airway obstruction on bronchoscopy and/or chest computed tomography. The remaining half of images were normal AP CXRs from children without TB disease. Three convolutional neural network (CNN) architectures - VGG16, ResNet50, and InceptionV3 - were evaluated and compared for their ability to classify pulmonary LBTB based on radiographic features.
Results:
A fine-tuned ResNet50 model achieved a statistically significant mean test sensitivity of 94.43% and a mean test specificity of 94.79% in identifying TB-positive cases. The model classified disease presence by detecting radiographic manifestations of LBTB affecting the trachea and main bronchi on CXR images.
Conclusion:
Deep learning models were successfully developed to detect significant airway obstruction secondary to pulmonary TB in children under 5 years of age. The best performing model, a customised ResNet50 architecture, achieved a statistically significant mean test sensitivity of 94.43% and a specificity of 94.79%. While these findings highlight the potential of CNN-based approaches for automated detection of paediatric LBTB, the limited sample size restricts definitive conclusions. Further validation using larger, multi-centre datasets is warranted.
