Evaluation of clinical image processing algorithms used in digital mammography

Area of Science:

• Medical Imaging
• Radiology
• Digital Mammography

Background:

• Screening reduces breast cancer mortality, but some cancers are missed.
• Digital mammography adoption necessitates evaluating image processing's role.
• Radiologist observations suggest image processing algorithms vary in perceived quality.

Purpose of the Study:

• To assess the impact of five manufacturer-recommended image processing algorithms on microcalcification cluster detection.
• To compare the performance of different algorithms using statistical analysis methods.

Main Methods:

• Retrospective collection of 200 digital mammograms (Siemens Novation DR).
• Insertion of simulated microcalcification clusters into half of the images.
• Processing of images with five algorithms (Agfa, IMS, Sectra, Siemens OPVIEW v2, Siemens OPVIEW v1).
• Radiologist evaluation using a five-point rating scale.
• Analysis of free-response data with jackknife free-response receiver operating characteristic (JAFROC) and receiver operating characteristic (ROC) methods.

Main Results:

• JAFROC analysis showed highly significant differences between image processing algorithms (p < 0.0001).
• Siemens OPVIEW2 and OPVIEW1 yielded the highest and lowest detection performances, respectively.
• ROC analysis also revealed significant differences, though less pronounced than JAFROC.
• JAFROC provided smaller confidence intervals compared to ROC analysis.

Conclusions:

• Image processing algorithms significantly influence the detectability of microcalcification clusters in digital mammograms.
• Objective measurements are crucial for manufacturers to select optimal image processing algorithms.
• This study underscores the importance of image processing in digital breast cancer screening.