Fig 1.
Prospective gating allows for synchronous image acquisition with breathing patterns.
Micro-CT acquisitions without (A) and with respiratory gating (B). Shown are coronal reconstructed tomograms (left) presenting a lung lesion (yellow arrow). Normalized breath signal (blue curve) and acquisition times (orange vertical lines) are shown for each gating technique (right). Note the sharper diaphragm and tumor margins in the gated acquisition.
Fig 2.
Process of adding simulated tumor to respiratory-gated and non-gated CT reconstructions.
Deformation fields between respiratory phases are used to place tumors in volumes corresponding to different respiratory phases. Once placed, each volume associated with its lung phase is forward-projected in cone beam geometry. Random combinations of respiratory-gated projections are used to simulate non-gated acquisitions. The projections from a single lung phase create gated acquisitions.
Fig 3.
Design of a pocket phantom for evaluation of respiratory gating.
The phantom was constructed using packaging foam to mimic the lung parenchyma in which a number of polyethylene spheres were inserted (A). The constructed pocket phantom was fixed to the thorax of free-breathing mice, allowing for uninhibited motion throughout the respiratory cycle (B). Examples of the micro-CT images of the pocket phantom (C) are shown in the axial (a.), coronal (b.), and sagittal (c.) view.
Fig 4.
Phantom assessment of the micro-CT scanner performance.
Shown is a slice and a line profile through the water phantom (A). The CT linearity was obtained by scanning a phantom containing 4 inserts with various concentrations of calcium hydroxyapatite (B). The spatial resolution was assessed with a wire phantom (C). Shown also is the image and the radial profile of NPS obtained in the water phantom (D). Finally, a coronal slice and a line profile of the multi-layer Defrise phantom is shown (E). No substantial cone beam artifacts are present.
Fig 5.
Respiratory gating improved the accuracy and precision of tumor volume measurements in the lung.
Volumes were calculated for a collection of tumors (A) in 3 positions per condition, including a non-gated image, a gated-image, and an image acquired post-mortem (standard). When accounting for variance, volumes from non-gated images were significantly different from post-mortem standards in half of cases, while gated image volumes did not differ significantly (B; *P<0.05, ANOVA). The resulting ANOVA table describes the deviation and difference in means between acquisition condition per tumor (C). Coefficients of variation were calculated for each condition and averaged for each acquisition type (D).
Fig 6.
QIBA-inspired profile for predicting tumor volumetry precision across variable tumor sizes in gated and non-gated images.
Shown is a table of sample anticipated tumor volumes for which the wCV and upper/lower 95% CIs was determined in both non-gated and gated acquisitions (A). Triplicate tumor measurements (plotted as mean ± standard deviation) from an independent cohort of mice were plotted against the calculated 95% CIs (red and blue bounding lines) (B), demonstrating the validity of the size-dependent confidence prediction for both non-gated (a) and gated (b) images.
Fig 7.
Demonstration of simulated lung tumor images compared to in vivo images of tumor-bearing lungs.
In vivo lung images from a tumor-bearing mouse demonstrate the difference between non-gated and gated images compared to post-mortem tumor detection (A). Simulated tumors are created from segmentations of real tumors, which are projected into the lungs of a healthy, free-breathing mouse (B). Simulated tumors mimic the appearance and location of tumors seen within in vivo images, as well as the effects of respiratory gating on tumor edge definition, but have the advantage of a binary label map which serves as “truth”. Real and simulated tumors are indicated with yellow arrows on in vivo and simulated acquisitions, respectively.
Fig 8.
Simulated tumors in CT of free-breathing mouse lungs demonstrate enhanced sensitivity, precision, and volume accuracy when identifying tumors in gated images compared to non-gated images.
Variably sized samples of measured simulated lung lesions are shown, including triplicate measures of binary label maps (truth), gated images, and non-gated images (A; ANOVA, *p<0.05). Bland-Altman plots of gated (B.a.) and non-gated (B.b.) tumor segmentations suggest agreement with segmentations of the binary label map, with a size-dependent bias noted in non-gated image segmentations. True positive rate (TPR; sensitivity), Miss rate (False negative rate, or lesions which the reader was unable to correctly identify), Positive predictive value (precision), False discovery rate (False positive rate), and F1 (harmonic mean of precision and accuracy) are shown for gated and non-gated images of simulated tumors of varied sizes (C). TPR as a function of simulated lesion size, with TPRs of 0.9 and 0.5 indicated (blue lines) (D).
Fig 9.
Retrospective gating improves tumor volume accuracy compared to non-gated images.
Representative images of prospectively gated, retrospectively gated, and non-gated images were acquired in a tumor-bearing mouse (A). SNR measurements were calculated from liver (red) and empty space (yellow) ROIs to determine mean SNR and variance (B). Three tumors per mouse (n = 5) were selected and segmented in triplicate under each scan condition, and results were compared with ANOVA (C). Bland-Altman comparisons to prospective gating were performed for retrospective gated and non-gated tumor analyses (D & E).
Fig 10.
The effect of gating on volumetric output and measurement variance.
Three pellets were measured three times in each acquisition condition, and volumes were calculated and compared to post-mortem measurements (A). Pellets were selected to mimic conditions regularly encountered during lung tumor analyses (B). Variance among measured volumes was greatest in non-gated images (red arrows in A), while gating resulted in precision similar to post-mortem measurements (C).