Abstract:
Pneumonia, one of the most common causes of morbidity and mortality globally, can be
caused by either bacterial or viral infection. With chest radiographs as the imaging
standard, the timely detection and diagnosis of the disease by experienced radiologists
and staff is necessary, especially in less-developed areas. Thus, this study aimed to
present an AI-based diagnostic tool that used the best optimizer and hyperparameter
values to identify and classify pneumonia from chest X-rays (CXR). Specifically, the
study aimed to identify the best AI-based diagnostic by evaluating its performance in
three (3) cases: between normal CXRs and pneumonia CXRs; between viral pneumonia
CXRs and bacterial pneumonia CXRs; and between normal CXRs, viral pneumonia
CXRs, and bacterial pneumonia CXRs. 1000 to 1500 CXR images were used in training
the AlexNet CNN model. In addition, different optimizers were evaluated across the three
classification cases to determine the model with the best-fit optimizer, which was the
RMSprop optimizer. After which, hyperparameter tuning was performed to further
optimize the model with image size = 150x150, dropout = 0.4, and batch size = 32. The
model was able to achieve an accuracy of 96.5% for classification case A, 81.5% for
classification case B, and 85.33% for classification case C. In conclusion, the model was
able to identify and classify pneumonia from CXRs in all three cases. Therefore, the
proposed model can be used as a supplementary diagnostic tool for pneumonia detection
and classification from CXRs.
