Automated Risk Prediction by Measuring Pneumothorax Size using Deep Learning

Shariful Islam, Hasin Rehana, Sayed Asaduzzaman, Syed Mobassir Hossen, Rabby Hossain, Touhid Bhuiyan, Muhammad Shahin Uddin, Nargis Akter

Research output: Contribution to conferenceConference paper

2 Citations (Scopus)


This research proposed an approach which takes images as a DICOM format from the Kaggle dataset named 'SIIM-ACR Pneumothorax Segmentation'. Preprocessed images were fed to popular Unet architecture with se_resnext50_32*4d architecture as backbone of the network, which could detect pneumothorax object on X-ray images as mask for semantic segmentation problem. Then, those mask images were post-processed for reducing noisy objects with a threshold value to identify the mask related to the pneumothorax region. Based on the mask images, percentage of the pneumothorax is calculated using C. Collins methods which also approximately determine the risk level of pneumothorax of a patient.
Original languageEnglish
Number of pages5
Publication statusPublished - 5 Jun 2020
Externally publishedYes
Event2020 IEEE Region 10 Symposium, TENSYMP 2020 -
Duration: 5 Jun 20205 Jun 2020


Conference2020 IEEE Region 10 Symposium, TENSYMP 2020


  • Deep Neural Network
  • Medical Image Processing
  • Pneumothorax
  • Quantification of Pneumothorax
  • Semantic Segmentation


Dive into the research topics of 'Automated Risk Prediction by Measuring Pneumothorax Size using Deep Learning'. Together they form a unique fingerprint.

Cite this