Early detection and speckle noise reduction in breast cancer ultrasound imaging using deep learning with enhanced ATTVNET 2.0.

Breast cancer is still among the main causes of death for women. Speckle noise, which usually degrades image quality, therefore affects the accuracy of the diagnosis. Early detection by ultrasound imaging is believed to be the most effective approach to increase survival rates. Conventional methods of noise reduction compromise detection accuracy by means of small structural detail preservation. This work presents an improved deep learning model called AttvNet 2.0 to manage the early detection of breast cancer as well as the reduction of speckle noise in ultrasound images of the condition. The proposed model comprises an attention-based convolutional neural network (CNN). This CNN is the most efficient method to extract features and lower noise. AttvNet 2.0 can maintain the structural details and increase the signal-to-noise ratio (SNR) by means of a multi-scale feature attention mechanism applied with a residual learning framework. The model's training and validation to be finished was aided by a breast cancer ultrasound dataset indicating improved accuracy, noise reduction, and lesion detection capacity. Under the conditions of noise reduction and lesion classification, the test results indicated that AttvNet 2.0 outperformed both U-Net and the Despeckle CNN (DS-CNN). It was 98.7% correct using AttvNet 2.0.