Computer simulations of case difficulty in digital breast tomosynthesis using virtual clinical trials

Area of Science:

• Medical imaging
• Computational modeling
• Radiology

Background:

• Virtual clinical trials (VCTs) are crucial for evaluating imaging technologies.
• Accurate simulation of patient data, including soft tissue lesions, is essential for VCTs.
• Model observer performance in VCTs depends on the quality of training data.

Purpose of the Study:

• To develop an efficient method for simulating and calibrating soft tissue lesions for VCTs.
• To ensure simulated lesions match the detectability threshold of virtual and human readings.
• To improve the reliability of VCTs in assessing imaging technology performance.

Main Methods:

• Simulated four mass models (ellipsoidal and spiculated) using composite voxels in anthropomorphic breast phantoms.
• Generated digital breast tomosynthesis projections and reconstructions.
• Evaluated Channelized Hotelling observers (CHOs) using simulated mass slices and calculated receiver operating characteristic (ROC) curves.

Main Results:

• Calibrated mass models achieved areas under the curve (AUC) within the 95% confidence interval of a previous human reader study (0.92-0.94).
• Results demonstrated that AUC varied significantly based on lesion shape, composition, and surrounding tissue density.
• The simulation method successfully matched the detectability threshold across different mass models.

Conclusions:

• Calibrating soft tissue lesions in VCTs to match human reader difficulty is critical for success.
• Lesion composition, shape, and size are key parameters for accurate VCT calibration.
• This method enhances the utility of VCTs for evaluating medical imaging systems.