Cervical Pre-cancerous Cell Image Classification System Using Histogram of Oriented Gradients and K-Nearest Neighbor Algorithms

Abstract

Cervical cancer is a dangerous disease, with more than 99% of which contain Human Papillomavirus (HPV), threatening women worldwide. The Global Burden of Cancer Study (Globocan) has recorded 36,633 cervical cancer cases, ranking second in Indonesia. Analysis of Pap smear results manually as an early detection effort possesses many weaknesses. Therefore, as an early detection step in diagnosing cervical pre-cancerous cell images, an artificial intelligence system is highly required to assist medical personnel in providing fast and accurate diagnostic evaluations. This study utilized 972 training image data and 108 testing image data, with a cervical pre-cancerous cells image classification system using Histogram of Oriented Gradients (HOG) algorithms for feature extraction and KNN machine learning for the classification system. The gray level in the contrasting images between the texture of the nucleus, cytoplasm, and background had different pixel and bit depth intensity values. Hence, HOG obtained bin orientation for each pixel in the cell. The cosine KNN model demonstrated the best matrix performance, acquiring classification results of 0.8 for accuracy, 0.8 for precision, 0.889 for recall, 0.846 for specificity, and 0.771 for f-score. Moreover, the training data generated an accuracy of 69.3% and the fastest training time of 4.2359s.