5TH International Congress on Technology - Engineering & Science - Kuala Lumpur - Malaysia (2018-02-01)

Chest X-ray Segmentation Using Cnn On Small Amounts Of High-variance Data

Lung fields and bone structures (ribs and clavicles) segmentation problem in chest X-Ray images is an important task as it improves the performance of other related tasks, such as ribs suppression and pathologies detection. In this paper, we used and analyzed the performance of convolutional neural networks (CNN) for the segmentation task. Specifically, we explore the case when only small amounts of high variance data is available for training. We used a dataset of 250 images containing both normal and pathological images from three different x-ray tools, in contrast to existing methods that train on thousands images. We present only raw CNN results without any pre- or post-processing. The results show that high variance affects the results and the accuracy is rather bad when testing set is different from the training one. However, CNN shows high accuracies on low-variance images even on small amounts of data.
Artem Kondyukov, Adel Zakirov, Adil Khan