Description

The primary objective of this task is identification and segmentation of chest radiographic images with pneumothorax.

Pneumothorax is usually diagnosed by a radiologist on a chest x-ray, and can sometimes be very difficult to confirm. An accurate AI algorithm to detect pneumothorax would be useful in a lot of clinical scenarios.

AI could be used to triage chest radiographs for priority interpretation, or to provide a more confident diagnosis for non-radiologists.

Introduction

Definition: Pneumothorax can be caused by a blunt chest injury, damage from underlying lung disease.

Data: Set of chest radiographic images in DICOM format(〜15.000 images) provided by Society for Imaging Informatics in Medicine

Dataframe

Data visualisation

Data augmentation

Machine learning

We used a U-net architecture with Adam as optimizer, BCE with logistics loss and ReLU activation function.
The network was trained on 256 images for 130 epochs.

Results

Image classification metrics

We used ROC and confusion matrix. Calculated them from segmented masks of images with two classes ("Healthy" and "Pneumothorax"). Their values are:
ROC

Confusion matrix

Image segmentation visualisation
Image segmentation metrics

Comparing the true labels with the ground truth gave following results:
Dice index: 0.043 +- 0.103
Intersection over union: 0.025 +- 0.065

Meet The Team

Team

Anton Frlan

  University of Rijeka
Team

Viktor Szvoreny

  University of Szeged
Team

Tadej Tomanič

  University of Ljubljana