Development of one channel-football formation neural network (OC-FFNet) for classification the breast ultrasound images

Abstract

Background: Breast cancer is a leading global health issue, with increasing incidence and mortality rates. In Thailand, it is the most common cancer among women, highlighting the need for better diagnostic methods. Traditional imaging techniques like mammography and ultrasound have limitations that hinder early detection. Recent advances in artificial intelligence (AI), particularly convolutional neural networks (CNNs), offer promising solutions for enhancing breast cancer detection in ultrasound images.

Objective: This study aims to develop a one-channel AI model for detecting breast cancer in ultrasound images, inspired by football formations to structure the CNN layers.

Materials and methods: The dataset comprises 18,000 breast ultrasound images categorized into normal, malignant, and benign cases. Data preprocessing involved image resizing, enhancement, and augmentation to address class imbalances. The proposed AI model, the One-Channel Football Formation Neural Network (OC-FFNet), was designed based on four distinct football formations: 4-3-3, 4-2-3-1, 4-4-2, and 5-4-1. Each formation guided the structuring of CNN layers, incorporating DenseNetbased modified dense blocks and transition layers. Model training was conducted with batch sizes ranging from 64 to 256 and epochs between 50 and 150. Performance evaluation metrics included accuracy, precision, recall, specificity, F1-score, false positive rate, and area under the curve (AUC).

Results: The models based on the 4-3-3 and 4-4-2 formations exhibited the highest classification performance, achieving an accuracy of 0.999, precision of 0.999, recall of 1.000, specificity of 0.999, F1-score of 0.999, and AUC of 0.999. The 4-2-3-1 model attained an accuracy of 0.963, while the 5-4-1 model achieved an accuracy of 0.968. Prediction times were consistent across all models, indicating computational efficiency. The findings suggest that formations with balanced positional distributions, such as 4-3-3 and 4-4-2, required fewer training iterations and larger batch sizes to achieve optimal performance.

Conclusion: The integration of football formation strategies into CNN architecture represents a novel approach to AI model design. The results indicate that strategically structured CNNs can improve breast cancer detection in ultrasound images.