A Sengupta*, M Lago, A Badano, Food & Drug Administration, Silver Spring, MD

• 
  A Sengupta

MO-E115-IePD-F8-4 (Monday, 7/11/2022) 1:15 PM - 1:45 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 8

Purpose: To integrate computational models of different detector technologies with the VICTRE in silico pipeline for breast imaging to evaluate the effect of detector technology on breast cancer detection.

Methods: We developed three detector models with varying image quality metrics. The DIR and DIR+ models replicate direct a-Se performance, while IND is based on indirect CsI technology. The DIR+ system also includes inter pixel crosstalk, resulting in lower resolution properties as compared to DIR. The same DM/DBT system geometry and x-ray acquisition parameters, modeled to mimic the Siemens Mammomat Inspiration system, were used to evaluate the three detectors. The VICTRE pipeline was used to simulate in silico patients with disease in form of spiculated masses or calcium oxalate clusters, generate DM/DBT images of the patients with the three detector models and finally interpret them using 2D/3D reader models. We analyzed the area under the ROC curves (AUC) for both imaging modalities and manifestations of breast cancer to characterize the clinical performance of the detectors.

Results: We found that the detector technology mainly affected the detection of microcalcifications, resulting in a drop in AUCs, especially in the DBT mode: on average, from 0.90 to 0.85 for the DIR and DIR+ systems respectively. For the DIR system, DM outperforms DBT for detection of microcalcifications and masses. For the DIR+ system, the average AUC difference value of -0.04 favoring DBT for the detection of microcalcifications.

Conclusion: The detector technology for the studied system parameters significantly affects the detection of microcalcifications. Detector technology does not significantly affect AUCs for DM but yields a significant drop in AUCs for DBT, especially for the DIR+ model. The AUC differences reveal that DM and DBT perform similarly for the DIR system, while DBT slightly outperforms DM for the DIR+ system.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email