Alternative Techniques for Breast Tumour Detection using Ultrasound

Area of Science:

• Medical imaging
• Artificial intelligence in healthcare
• Biomedical signal processing

Background:

• Ultrasound imaging is crucial for diagnosing internal organ diseases.
• Image noise significantly hinders the interpretation of ultrasound scans.
• Accurate segmentation and detection of lesions are vital for effective diagnosis.

Purpose of the Study:

• To compare supervised machine learning approaches for lesion modeling in ultrasound images.
• To utilize unsupervised machine learning for tumor segmentation and detection in breast ultrasound.
• To evaluate the effectiveness of different feature extraction and machine learning techniques.

Main Methods:

• Application of supervised machine learning algorithms to train lesion models.
• Integration of unsupervised machine learning for image segmentation and tumor detection.
• Experimentation using two synthetic and one real-world breast ultrasound dataset.
• Comparative analysis of various feature extraction methods, including Frost Filter and Quick Shift.

Main Results:

• The Frost Filter combined with the Quick Shift algorithm yielded the highest system performance.
• Machine learning approaches demonstrated potential in overcoming noise-related interpretation challenges.
• Successful segmentation and detection of tumors were achieved across different datasets.

Conclusions:

• The combination of Frost Filter and Quick Shift offers a promising approach for enhancing breast tumor detection in ultrasound.
• Supervised and unsupervised machine learning techniques can significantly improve the accuracy of medical image analysis.
• Further research can optimize these methods for broader clinical application in diagnostic imaging.