(G-249) Impact of Object Size on Magnetic Particle Imaging (MPI) Quantification Accuracy

Each calibration curve and model set were filled with a serial dilution of the commercially available SPION, ferucarbotran, and deionized water. The disc models filled with 1μg, 10μg and 100μg of iron were imaged in 2D high sensitivity (HS), high resolution (HR), and high sensitivity/high resolution (HSHR) scan modes. Sepatate calibration curves were constructed using different volumes, specifically 1μL in capillary tubes, 16μL in Eppendorf tubes, and 415μL in disc models F. These calibration curves developed a relationship between iron mass and total signal. The 1μL calibration curve and sphere models filled with 10μg of iron were imaged in 3D in HR to identify any differences from 2D to 3D imaging.

Each MPI scan was analyzed using 3D Slicer. The region of interest was identified and then thresholded to a lower limit of half the maximum voxel intensity signal. From this, the mean voxel intensity signal within the ROI was multiplied by the number of voxels to calculate the ‘total signal’. The averages of the total signals were plotted against iron mass for each calibration curve type, generating a trendline equation relating total signal and iron mass. Estimations of the iron mass within each model were then developed using these equations.

Results and Discussion

The results of this study show iron quantification errors within MPI due to its finite resolution. Figure 2 shows the percent errors of the iron estimations of the 10µg disc set with respect to the true amount of iron filled within each model, considered ground truth. The capillary tube and Eppendorf tube calibration curves typically caused overestimation of the iron mass, while the high volume calibration curve caused underestimation of the iron mass, as shown by the negative percent error. These findings indicate that volume of both calibration standards and models plays a significant role in the accuracy of iron quantification for MPI. Figure 2 also demonstrates a relationship between the calibration curve and model volumes. As the volume of the models approach that of the calibration curves, the iron estimation becomes more accurate. It is this trend that led to more accurate quantifications of models A-C using the capillary and Eppendorf tube models, and more accurate quantifications of model D-F using the high volume calibration curve. Additionally, the impact of the resolution error appears to vary between scan modes, with HS producing the most accurate results.

The iron estimations of the 3D samples were evaluated with respect to the 2D models. Figure 3 shows the percent errors of the 10µg sphere models imaged in the MPI in 3D, HR mode compared to the 10µg disc models imaged in the MPI in 2D, HR mode. Notably, the errors experienced in 3D images are reduced relative to the 2D images, likely due to a more accurate spatial resolution provided within the 3D images.

Conclusion

These results confirm the existence of a resolution related error impacting MPI quantification. The extent of the error varies between scan modes (HS, HR, and HSHR), with HS being the most accurate of the three. The error also varies between 2D and 3D images, with 3D images resulting in more accurate iron quantification, likely due to reduced averaging of the SPIONs. This information introduces the opportunity to study correction methods to improve MPI accuracy.