Classification of breast tumor ultrasound images with artificial intelligence techniques and statistical methods

Breast cancer remains a significant global health concern, where early diagnosis is crucial for improving patient outcomes. Ultrasound imaging, a non-invasive diagnostic tool, provides valuable insights into tumor characteristics. However, it presents challenges in distinguishing between benign and malignant tumors due to overlapping visual features. Artificial intelligence (AI) techniques have shown promise in enhancing classification accuracy within medical imaging., This study investigates the effectiveness of four AI models Artificial Neural Networks (ANN), Support Vector Machines (SVM), Decision Trees (DT), and k-Nearest Neighbors (k-NN) in classifying breast tumors from ultrasound images. Therefore, a dataset of 300 ultrasound images, evenly split between 150 benign and 150 malignant cases, was used to evaluate each AI model. Consequently, results indicate that ANN achieved the highest accuracy at 87.78%, followed by SVM 86.67%, DT 83.33%, and k-NN 72.22%. Specificity was highest for SVM at 95.56%, while sensitivity peaked for ANN at 86.67%. ANN also yielded favorable results in positive and negative predictive values, at 88.64% and 86.96%, respectively. The Area Under the Curve (AUC) values demonstrated strong performance for both ANN 87.80% and SVM 86.70%, whereas k-NN showed the lowest performance across most metrics. Finally, this study further reveals that malignant tumors exhibit greater heterogeneity and irregularity in shape compared to benign tumors, while benign tumors tend to display a greater degree of roundness. Then, the findings suggest that ANN and SVM provide robust accuracy and reliability for breast tumor classification, outperforming DT and k-NN. Keywords: Artificial neural networks, Breast tumor, Decision trees, Image Processing, k-nearest neighbors, Support vector machines, Ultrasound