Implementation of CNN Method with Otsu Thresholding Preprocessing for Pneumonia Detection

Abstract

Pneumonia is a lung infection requiring rapid diagnosis to prevent fatal complications, yet X-ray image quality often hinders manual detection accuracy. This study proposes a hybrid approach using a Convolutional Neural Network (CNN) optimized with Otsu Thresholding for lung area (Region of Interest) segmentation. Experiments were conducted on 1,840 images from a secondary dataset. Evaluation results demonstrate a highly balanced and superior model performance, achieving 96% Recall, 91% F1-Score, and 96% Accuracy. The alignment between accuracy and recall values indicates that the model possesses equally good sensitivity and specificity in detecting both positive and negative cases. These findings prove that Otsu pre-processing effectively assists the CNN in focusing on pathological features, making this method a promising automated diagnostic solution.