A Phantom Study for Assessing the Effect of Different Digital Detectors on Mammographic Texture Features

Area of Science:

• Medical Imaging
• Radiology
• Biomedical Engineering

Background:

• Digital mammography (DM) is a standard breast cancer screening tool.
• Research using DM datasets requires understanding detector-specific image variations.
• Inconsistent detectors can affect the reliability of texture analysis in breast imaging.

Purpose of the Study:

• To compare the impact of two digital mammography detectors (GE 2000D and DS) on texture analysis.
• To evaluate the effectiveness of image normalization techniques in mitigating detector-induced differences.
• To assess detector effects on various texture features derived from raw and processed DM images.

Main Methods:

• Acquisition of digital mammography images in Cranio-Caudal view using GE 2000D and DS detectors.
• Texture feature generation for both raw and post-processed images.
• Comparison of image intensity profiles and texture features, including analysis of cumulative distribution function (CDF) curves.
• Application of z-score normalization to assess its impact on detector differences.

Main Results:

• Significant inherent differences were observed between the GE 2000D and DS detectors.
• CDF curves showed distinct scaling and shifting factors for raw and processed images, respectively.
• Z-score normalization effectively reduced differences in grey-level intensity and histogram-based texture features.
• Co-occurrence and run-length texture features remained different even after normalization, indicating persistent detector effects.

Conclusions:

• Digital mammography detector choice introduces inherent image variations affecting texture analysis.
• Z-score normalization partially mitigates detector-related artifacts, particularly for intensity and histogram features.
• Further methods are needed to address persistent differences in spatial texture features caused by detector variations.