Patient-specific organ dose and in-vivo image quality assessment in clinical CT

Abstract

Purpose

To develop and characterize individualized dose and quality measures at organ level compared to their generic counterparts across a clinical CT dataset.

Materials and methods

The study included 9801 chest-abdomen-pelvis and abdomen-pelvis CT exams (7,763 patients, mean age, 56 ± 17 years; 4113 women) representing 20 unique protocols. For each exam, patient-specific organ dose of all radiosensitive organs was estimated using a validated method by generating personalized computational phantoms and Monte Carlo simulations. Effective dose ($E_{\mathrm{OD}}$) was calculated by weighted sum of the organ doses. Liver dose, $O{D}_{\mathrm{liver}}$, noise in the liver, $N_{\mathrm{liver}}$, and observer model detectability, $d\prime$, were assessed within the liver as examples of individualized, organ-based image assessment measurements. The organ-based measurements ($O{D}_{\mathrm{liver}}$, $E_{\mathrm{OD}}$, and $N_{\mathrm{liver}}$) were compared to their generic counterparts: ssize-specific ddose estimates (SSDE), effective dose based on dose length product ($E_{\mathrm{DLP}}$), and whole-body noise ($N_{\mathrm{global}}$), respectively.

Results

Generic dose values were substantially higher than individualized estimates for SSDE vs. $O{D}_{\mathrm{liver}}$ (median of all exams: 51.2 %, p < 0.001) and $E_{\mathrm{DLP}}$ vs. ${\mathrm{ED}}_{\mathrm{OD}}$ (median: 41.0 %, p < 0.001). $N_{\mathrm{global}}$ was generally lower than $N_{\mathrm{liver}}$ (median: −7.2 %, p < 0.001). The correlation relationships of $E_{\mathrm{OD}}$ and $d\prime$ were substantially varied ($R^2$ range: 0–0.5) for different patient sizes and scan parameters.

Conclusions

Demonstrated across a population of exams, individualized organ-based measurements of dose and quality are feasible. Generic measures cannot fully represent individualized organ-based values. The correlation relationships between individualized dose and image quality values varies for different vendors and protocols, implying imaging optimization is best when done semi-independently for each factor using individualized measurements.