1. Collected at least 500 chest x-rays with different categories of positive ILOs.
2. Evaluated our automated pneumoconiosis detection tool’s sensitivity and specificity using these additional data. Made necessary modification to the design of the tool, if needed.
3. Developed, trained, and evaluated an automated pneumoconiosis classification tool using the additionally collected data and synthetic images learnt from the real X-ray image data. Such a tool is able to predict an ILO Classification System based category of severity of pneumoconiosis, and, possibly, the correct shape and size of parenchymal opacities that are main signs of pneumoconiosis in chest x-rays.
4. Developed a user-friendly front end to the detection and classification tools and delivered the software to CSH. This pilot software is used retrospectively on 5% of images routinely collected by CSH and evaluated against expert radiologists’ examination results. The feedback from the pilot study will be used to further improve functionality and performance of our automated tools