Hybrid Mixer Framework Enhances Breast Ultrasound Image Segmentation Accuracy

Strategic Insights -

A new research study has unveiled a machine learning architecture that significantly improves breast ultrasound image segmentation accuracy. The novel framework, known as HMA-Net, uses a combination of ConvMixer and ConvNeXT modules, along with convolution-enhanced multihead attention, to effectively capture local and global contextual features. This enables the model to accurately detect and localize tumors, reducing the mortality rate associated with breast cancer. According to the research, the proposed model has demonstrated exceptional performance on two datasets, achieving a Jaccard index of 98.04% and 94.84% and a Dice similarity coefficient of 99.01% and 97.35% on the BUSI and BrEaST datasets, respectively.

Key Takeaways:

  • The HMA-Net framework is a novel machine learning architecture that utilizes ConvMixer and ConvNeXT modules, along with convolution-enhanced multihead attention, to enhance breast ultrasound image segmentation accuracy.
  • The model has been designed to capture local and global contextual features, enabling accurate detection and localization of tumors.
  • Comprehensive experiments conducted on two datasets, BUSI and BrEaST, demonstrated the model's exceptional performance, achieving a Jaccard index of 98.04% and 94.84% and a Dice similarity coefficient of 99.01% and 97.35%, respectively.
  • The ConvMixer and ConvNeXT modules are integrated with convolution-enhanced multihead attention, which enhances the model's ability to capture local and global contextual information.
  • The proposed model has the potential to significantly reduce the mortality rate associated with breast cancer by enabling early and accurate detection.
  • The research was conducted by L. Jani Anbarasi and the School of Computer Science and Engineering at the Vellore Institute of Technology in Chennai, India.

Statistics:

  • Jaccard index: BUSI dataset (98.04%), BrEaST dataset (94.84%)
  • Dice similarity coefficient: BUSI dataset (99.01%), BrEaST dataset (97.35%)
  • datasets used: BUSI and BrEaST
  • modules used: ConvMixer, ConvNeXT
  • attention mechanism: convolution-enhanced multihead attention
  • loss function: combined binary cross entropy and dice loss

Sources:

  • HMA-Net: a hybrid mixer framework with multihead attention for breast ultrasound image segmentation (HMA-Net). Frontiers in Artificial Intelligence, 2025;8:1572433.
  • NewsRx LLC. School of Computer Science and Engineering Reports Findings in Artificial Intelligence (HMA-Net: a hybrid mixer framework with multihead attention for breast ultrasound image segmentation). Robotics & Machine Learning. July 14, 2025; p 847.