Optimization of micro-CT for preclinical imaging of lung metastases
There are many aspects of micro-CT imaging that are important for optimal performance, independent of motion such as calibration, image acquisition, image reconstruction and post-processing parameters (Table 1). We have built multiple micro-CT systems[38, 62], thus we master these tasks[39, 63-65].

Table 1: Optimization of micro-CT imaging

Micro-CT Calibration includes: i) X-ray source conditioning (10-15 mins) ii) dark/light (D/L) projection acquisition required for log data normalization and iii) geometric calibration.
D/L (dark/light) normalizations are used to obtain line integrals required for reconstructions and are acquired when X-ray source is turned either off (D) or on (L) and the field of view is empty. 40 images are typically considered sufficient, but we will carefully assess image quality as a function of their numbers.D/L normalizations are affected by pixel binning, field of view (FOV), X-ray exposure and are sensitive to system temperatures, which is why they should be run daily as part of quality assurance. X-ray beam filtration affects D/L normalization and the exposure time. A filter made of aluminum (0.5 mm) is common. No data saturation should occur when acquiring light images.

Geometric calibration: Precise micron-scale reconstruction, requires accurate knowledge of the system geometry. Commercial scanners are geometrically calibrated at the factory, but require adjustments over time. Poor geometric calibration results in inaccurate reconstruction. Geometric calibration generally uses a reference phantom.

 Image Acquisition: High-resolution scans should cover a 360º rotation with at least 360 projections. The estimated scan time will be proportional to the number of projections acquired and the exposure time for each projection.

 Spatial Resolution: We have shown that the biological variability in gating across multiple heart beats limits in-plane spatial resolution to ~100 µm for in vivo lung imaging [66]. We will aim for ~100 µm spatial resolution at the lowest practical radiation dose.

Exposure Settings:  For typical micro-CT systems with micro-focus x-ray tubes, the accelerating voltage (kVp) ranges up to 80 kVp, where we also aim to scan. But lower voltages can be also selected for special purposes such as achieving higher contrast of tissue or other low-density material. The typical micro-focus X-ray tube anode current ranges up to 500 μA. For a given operating voltage, higher anode current provides higher flux and higher SNR in projection images with shorter exposures. Short exposures are required for respiratory gating.

Gating: Assuming that the mouse’s respiratory rate is 2 s⁻¹ (a normal rate for a warm animal under proper anesthesia), the expiratory period is ~250 ms per breath. Unlike in clinical chest CT, which is done in a single breath hold, preclinical projection data must be acquired over many breaths, requiring respiratory gating. Respiratory gating can be performed prospectively or retrospectively. In prospective respiratory gating, a single respiratory phase (e.g. end expiration) can serve well to assess lung nodules. We were first to develop and implement combined cardiac and respiratory gating for micro-CT [38], which provides the highest possible lung image quality.

Image Reconstruction and Post-Processing: The most used reconstruction algorithm for cone beam micro-CT is filtered back projection (FBP) based on Feldkamp’s algorithm [67]. FBP algorithms use filters. The most common are: i) Ramp filter which, retains high frequency information and minimizes low-frequency blurring; ii) Shepp-Logan filter which suppresses some high-frequency noise in the projections and provides higher resolution images; iii) Hamming filter  produces smoother images. We expect that Shepp-Logan would be the preferred filter since it compromises between lower noise and sufficient detail. Other algorithms are implemented on some commercial scanners (e.g. Inveon Siemens) such as Ordered Subsets Expectation Maximization (OSEM) [68] and Maximum A-Posteriori (MAP) estimation [69]. OSEM and MAP are iterative methods and thus slower than FBP algorithms. However, they can produce higher quality images compared to Feldkamp. Although we consider FBP as the most robust algorithm for best practices in micro-CT reconstruction, we will compare Feldkamp, OSEM, and MAP in a task based approach, where the task will be to accurately identify and measure lung nodules in simulations and real experimental data.

Beam-hardening corrections lessen image artifacts caused by low-energy X-rays that are attenuated at a higher rate than higher energy X-rays. These artifacts can be particularly significant in the lungs which are surrounding by highly attenuating ribs. Beam-hardening correction is implemented on some commercial scanners (e.g. Inveon Siemens) and typically linearizes attenuation measurements relative to a set of calibration measurements [70]. Scatter corrections will also be tested[71]. Notably, such corrections typically increase noise in the resultant reconstruction, so we will test beam hardening corrections specifically for the task of lung nodule identification and measurement in mice, allowing us to make recommendations.

  Hounsfield Scale Calibrations: In CT, the image values are typically expressed in Hounsfield units (HU). HU calibration requires attenuation measurements in air and water, which can be made using common software such as ImageJ (https://imagej.nih.gov). Ideally, phantom measurements should be used for reproducibility and for uniformity between CT data sets. Knowing that the CT HU values for air should be -1000 and for water 0, the procedure determines the slope and the intercept for corrections. The water values after HU calibration should ideally be 0 and not exceed +/- 50 HU. The HU calibrations should be performed for the weekly QC or after any acquisition changes. Micro-CT should produce images with acceptable levels of noise and doses.

  Image Noise: Relative to clinical CT, preclinical micro-CT has much higher spatial resolution requirements (~100 μm). For a typical imaging protocol (e.g. 80 kVp, 500 μA, 0.5 mm filtering, 512 projections, 100 µm³ _­­voxels) the standard deviation of noise in water should be between 50-70 HU. An unstable X-ray flux could result in higher noise levels or artifacts in the reconstruction. We recommend a QC procedure every week to assess the stability of the X-ray flux and the image uniformity by performing a HU Calibration.

Radiation Dose:  The radiation dose associated with micro-CT scan should be < 0.07 Gy per mouse. This is ~75-100 times less than LD50/30 lethal dose (5-7 Gy) in mice [72] and ~285 times less than the radiation therapy dose. At this low levels, the micro-CT imaging dose will not influence lung nodule developments[73].

Performance Analysis and Cross-Validation
To assess image quality, we will use the commercially available performance evaluation micro-CT phantoms (www.simutec.com ). The phantom comes with an automated evaluation software (Model vmCT-SOFT). We will assess multiple image-quality parameters of the CIVM available micro-CT systems. We will characterize their performance quantitatively, in terms of resolution, geometric accuracy, CT number accuracy, linearity, and noise and image uniformity using the various modules contained by this phantom.  We will design the imaging protocols for all scanners to provide similar image quality. Next, we will follow the metrology standards proposed by QIBA to measure bias, linearity, precision, repeatability and reproducibility of our CT measurements [75]. Typically  lung nodules in the mouse range from 0.5 to 4.0 mm in diameter [76].  We hypothesize that the smallest lesion that we will be able to measure with micro-CT will be around 0.7 mm, but this will be carefully investigated. A precision made phantom similar to the CT “pocket phantom”, [77] will be used as a reference to quantify errors. The phantom will have 4 precision-manufactured spheres simulating lung nodules (diame. 0.5, 0.7, 1.3 and 1.3 mm). Mimicking the attenuation differences between lung nodules and normal lung tissue, the spheres will be made from Teflon and embedded in a urethane block .