Pneumothorax Detection and Classification on Chest Radiographs using Artificial Intelligence

Author(s): Tejas Haritsa V K, Naveen Raju S G, Kishore Rajendra, Arjun Kalyanpur, Pallavi Rao


In recent years, Computer Aided Diagnosis (CAD) systems have been designed for the detection of lung space anomalies. Pneumothorax is an abnormal collection of air in the pleural space between the lung and the chest wall that can result in partial or complete lung collapse [1]. This is a medical emergency in which quick detection and timely intervention can be life-saving. Pneumothorax detection on chest radiographs is important & may be facilitated with the help of image processing and deep learning algorithms. In this study we aim to evaluate the performances of two artificial intelligence systems in detection of pneumothorax on chest radiographs. The AI system was trained on open source datasets obtained from the US National Institutes of Health (NIH) [2], Society for Imaging Informatics in Medicine (SIIM) [3][4] & private datasets. Two unique approaches were used, one involved processing high-resolution complete images of size 1024x1024px and other involved feeding medium resolution images in portions (segments), each of size 448x448px. The trained AI systems was trained with binary mask as ground truth evaluated by a team of radiologists where, the segmental approach yielded a dice coefficient of 0.72, sensitivity of 0.986, specificity of 0.95, accuracy of 0.9683 and with Area Under Receiver Operating Characteristic Curve (AUROC) of 0.95, while the full image approach yielded an accuracy of 0.9417, dice coefficient of 0.865, sensitivity of 0.9084, specificity of 0.9510 with AUROC of 0.93.

The Chartered Data Scientist Designation

Achieve the highest distinction in the data science profession.

Explore more from Association of Data Scientists

Become ADaSci Chapter Lead

As a chapter lead, you will have the opportunity to connect with fellow data professionals in your area, share knowledge and resources, and work together to advance the field of data science.