Pneumonia is an inflammatory disease that attacks the lungs. Infection with this disease causes fluid to appear in one or both lungs. One way to establish a diagnosis of pneumonia is by analyzing chest X-ray images. However, the use of this method needs to be improved in terms of accuracy. This was conveyed by Prof. Chomsin Sulistya Widodo, S.Si., M.Si., Ph.D. in his scientific oration for the inauguration of the professor, on Tuesday (25/2).
He developed a digital analysis model called UBNet_v3 which can identify and classify pneumonia caused by bacterial, viral or Covid infections. “Increased detection speed and accuracy can help radiologists,” he explained.
The development of UBNet_v3 was carried out by combining the UBNet_v1 and UBNet_v2 models. The performance of UBNet_v3 to detect and classify non-covid viral pneumonia and covid pneumonia is very satisfying where all categories, namely accuracy of 99.6% are accurate, precision of 100%, sensitivity of 99.45% and F1 score of 99.72%.
Complementing the results of the developed model, a GUI-based display system has been successfully created. System performance testing was carried out using a laptop with a Ryzen 5 processor and 8GB RAM, without using an external GPU. The average time required to perform the analysis was 1.21 seconds for each image.
“The use of a simple CNN architecture with a total of 7 layers consisting of 5 convolution layers and 2 ANN layers. Architecture with a small number of layers can lighten the computational load. Meanwhile, other CNN architectures use layers between 16 and 71, as shown in the reference table. The performance of UBNet_v3 architecture is very satisfying where all categories, namely accuracy of 99.6% are accurate, precision of 100%, sensitivity of 99.45% and F1 score of 99.72%, “explained the lecturer at the Faculty of Mathematics and Natural Sciences.
Chomsin is a professor in the field of Medical Physics and Radiology who is the 34th active professor at FMIPA, as well as the 414th professor produced by Universitas Brawijaya. (VQ/ UB PR/ Trans. Iir)
