Vision-Transformer-Based Transfer Learning for Mammogram Classification.

Area of Science:

• Medical imaging
• Artificial intelligence in healthcare
• Oncology

Background:

• Accurate breast mass identification in mammograms is vital for early breast cancer diagnosis.
• Distinguishing benign from malignant breast lumps in early stages remains challenging.
• Convolutional Neural Networks (CNNs) have advanced mammogram analysis but have limitations in scope and computational efficiency.

Purpose of the Study:

• To develop and evaluate a novel transfer learning technique utilizing vision transformers for classifying breast mass mammograms.
• To assess the performance of vision transformers in the medical imaging domain for breast cancer detection.
• To improve the accuracy and efficiency of early breast cancer diagnosis.

Main Methods:

• Development of a transfer learning model based on vision transformers specifically for breast mass mammogram classification.
• Comparison of the vision transformer model against CNN-based transfer learning models and models trained from scratch.
• Evaluation of model performance using the area under the receiver operating curve (AUC).

Main Results:

• The developed vision transformer transfer learning model achieved an area under the receiver operating curve of 1 ± 0.
• This performance significantly outperformed existing CNN-based transfer learning models.
• The model also demonstrated superior results compared to vision transformer models trained from scratch.

Conclusions:

• Transfer learning with vision transformers shows exceptional promise for accurate breast mass classification in mammograms.
• This technique offers a potential advancement for improving early breast cancer diagnosis in clinical settings.
• Vision transformers represent a powerful tool for medical image analysis, overcoming limitations of traditional CNNs.