수퍼픽셀 기반의 합성곱 신경망 성능 평가

초록

Recently, the integration of Robotic Process Automation (RPA) and Artificial Intelligence (AI), known as Intelligent Process Automation (IPA), has made a significant impact across various industries. However, its application in medical image analysis remains in the early stages. This study proposes a convolutional neural network (CNN)-based image classification system for diagnosing pneumonia from chest X-ray images and compares the performance of models using original image data versus superpixel-based representations. Experimental results show that the superpixel-based CNN outperforms the original image-based model in terms of test accuracy, recall, and F1-score, particularly demonstrating higher recall for the pneumonia class, indicating its effectiveness in disease detection. Furthermore, the superpixel-based model exhibits greater stability in terms of training convergence speed and validation accuracy, suggesting superior generalization performance. These findings imply that the superpixel-based approach preserves structural features of medical images more effectively and enables robust learning against noise. As a meaningful case of applying IPA to medical image analysis, this study highlights the potential of improving diagnostic accuracy while contributing to enhanced operational efficiency, reduced diagnostic costs, and decreased patient wait times in medical institutions.

키워드