A Semi-Supervised Method for Tumor Segmentation in Mammogram Images.

Area of Science:

• Medical Imaging
• Machine Learning
• Computational Biology

Background:

• Breast cancer is a prevalent disease in women, with mammography playing a crucial role in diagnosis and treatment.
• Machine learning, particularly pixel-based segmentation, is increasingly utilized for tumor identification in mammograms.
• Traditional supervised methods require extensive labeled data, which is costly and time-consuming to obtain, while unsupervised methods may lack accuracy.

Purpose of the Study:

• To develop an accurate and efficient semi-supervised learning method for tumor segmentation in mammogram images.
• To address the limitations of supervised and unsupervised learning approaches in medical image analysis.
• To improve the performance of tumor segmentation using a cotraining algorithm with reduced feature sets.

Main Methods:

• A semi-supervised learning approach was employed for pixel-based tumor segmentation.
• Static and gray level run length matrix features were extracted for each pixel.
• Fisher discriminant analysis (FDA) was used for feature reduction, followed by a cotraining algorithm integrating Support Vector Machine and Bayes classifiers on the MIAS dataset.

Main Results:

• The proposed semi-supervised method demonstrated superior performance in tumor segmentation compared to existing supervised methods.
• The integration of feature reduction and cotraining algorithms improved segmentation accuracy.
• The method effectively utilizes limited labeled data for robust tumor identification.

Conclusions:

• The developed semi-supervised method offers a promising alternative for accurate and efficient tumor segmentation in mammography.
• This approach overcomes the data dependency of supervised methods and the potential performance issues of unsupervised methods.
• The findings suggest a significant advancement in automated breast cancer detection through machine learning.