Author: Xiaoyu Duan, Xinyu Hu, Runqiu Li, Xiang Li 👨🔬
Affiliation: Dukekunshan University, Medical Physics Graduate Program, Duke Kunshan University 🌍
Purpose:
Accurate detection of small-sized microcalcifications (μCalcs) (< 500 µm) is critical for early breast cancer diagnosis, requiring optimal imaging systems and reconstruction algorithms. However, the distinct image quality characteristics, particularly for high-frequency signals, produced by linear and no-linear reconstruction algorithms, complicate the generalization of digital breast tomosynthesis (DBT) system optimization for small-sized μCalc detection across different algorithms. Therefore, this study systematically evaluates DBT system optimization strategies for μCalcs detection improvement under various reconstruction methods, including linear and non-linear algorithms, using an in silico experimental pipeline.
Methods:
DBT projections of a digital anthropomorphic breast phantom inserted with 120-μm μCalc clusters were simulated under various imaging conditions: with/without focal spot motion (FSM), uniform/non-uniform angular dose distribution (ADS), and AMFPI/CMOS detector. Volumes were reconstructed with linear (filtered back projection (FBP)) and non-linear (simultaneous algebraic reconstruction technique (SART)) algorithms. The evaluation metrics are the signal-to-noise ratio (SNR) of the filtered channel observers and the area under the receiver operating characteristic curve (AUC) derived from multiple-reader, multiple-case analysis.
Results:
For both reconstruction algorithms, removing FSM enhanced μCalc sharpness and increased SNR and AUC by 49.5% and 6.4% for SART and 5.5% and 1.8% for FBP. Conversely, using non-uniform ADS showed no benefit for μCalc detection in SART-generated images but increased SNR and AUC by 62.8% and 10.2%, respectively, for FBP. Utilizing a CMOS detector demonstrated the most substantial enhancement in μCalc detectability across both reconstruction algorithms, improving SNR and AUC by 39.0% and 2.1% for SART and 89.2% and 12.8% for FBP.
Conclusion:
In conclusion, our study indicates that the improvement brought by image acquisition optimization varied under different reconstruction algorithms. The choice of the reconstruction algorithm should be considered when determining the priority of DBT system optimization.