Predicting Breast Tumor Malignancy Using Deep ConvNeXt Radiomics and Quality-Based Score Pooling in Ultrasound...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early breast cancer detection is crucial for reducing mortality rates.
• Ultrasound imaging (US) aids diagnosis, especially in dense breasts.
• Current computer-aided diagnosis (CAD) systems using single US images have limited accuracy due to image quality and tumor variability.

Purpose of the Study:

• To propose a deep-learning-based radiomics method utilizing breast US sequences for enhanced breast cancer diagnosis.
• To address limitations of single-image CAD systems, including tumor variability and image noise.

Main Methods:

• Radiomic features extraction using a deep learning network (ConvNeXt), trained with a vision transformer approach.
• An efficient pooling mechanism to integrate malignancy scores from US sequence frames, considering image quality statistics.
• Visual interpretation of results.

Main Results:

• The proposed method demonstrates competitive performance compared to existing CNN-based approaches.
• Ablation studies and experimental results validate the effectiveness of the deep learning radiomics method.
• The approach shows promise in overcoming challenges associated with US image analysis for breast cancer.

Conclusions:

• The deep-learning-based radiomics method using breast US sequences offers an advanced approach for breast cancer diagnosis.
• This method shows potential for improving diagnostic accuracy, especially in complex cases with dense breasts.
• The integration of ConvNeXt and a novel pooling mechanism provides a robust framework for analyzing ultrasound sequences.