K-Means Clustering and Classification of Breast Cancer Images Using Histogram of Oriented Gradients Features and Convolutional Neural Network Models: Diagnostic Image Analysis Study

Background Breast cancer has proven to be the most common type of cancer among females around the world. However, mortality rates can be reduced if it is diagnosed at the initial stages. Interpretation made by an expert is required by conventional diagnostic tools such as biopsies and mammograms, and this interpretation can be erroneous. Artificial intelligence has increasingly been used to detect and classify breast cancer. Yet, the acquisition of impressive reliability and accuracy continues to be challenging with these automated systems. Objective This study aimed to develop an innovative hybrid technique for the classification of breast cancer images involving unsupervised analysis by K-means clustering, feature extraction using Histogram of Oriented Gradients (HOG), and classification of images through a convolutional neural network (CNN). Methods This study used a dataset of 2788 breast cancer images categorized into benign (n=1480) and malignant (n=1308) classes. The proposed hybrid method included three stages: (1) unsupervised clustering using K-means to group visually similar features; (2) feature extraction using Histogram of Oriented Gradients (HOG) to capture texture and shape patterns; and (3) classification using a CNN trained on the extracted features. The model's performance was evaluated using standard metrics such as accuracy, precision, recall, and F 1 -score. Results The CNN achieved a classification accuracy of 98%, with precision, recall, and F 1 -score values of 0.98 for both benign and malignant cases. K-means clustering revealed distinct groups corresponding to benign and malignant tumors, indicating natural separability in the image data. Conclusions The combination of HOG feature extraction and CNN-based classification demonstrated high performance in breast cancer detection. The model demonstrates potential utility for automated diagnosis, with possible clinical applications to assist radiologists in identifying malignant tumors more efficiently. Future research will explore additional imaging modalities and real-world clinical testing.