Patients treated at the Toronto General Hospital-ILD clinic, a tertiary academic referral clinic, were approached for study enrolment between July 2012 and April 2013. Oral and written consent was recorded when the patients were approached while visiting the clinic for regularly scheduled appointments. Patient consent records were stored securely with their study records. Adult patients (≥18 years of age) with a diagnosis of IPF, idiopathic NSIP, or connective tissue disease (CTD)-ILD and who required a thoracic CT at their visit were included. Patients with a history of pneumotoxic medication use, environmental exposures, granulomatous lung disease (i.e. sarcoidosis) or other idiopathic interstitial pneumonias were excluded from this study. A diagnosis of IPF was based on criteria outlined by American Thoracic Society guidelines.5,6 A rheumatologist evaluated all patients to confirm the presence of a CTD (rheumatoid arthritis, systemic sclerosis, idiopathic inflammatory myositis, Sjogren's syndrome, mixed connective tissue disease, systemic lupus erythematosus, or undifferentiated connective tissue disease).

Ethics approval was obtained prior to patient enrolment from both the BLINDED FOR REVIEW Research Ethics Board and the BLINDED FOR REVIEW Research Ethics Board.

Patients were scanned in the prone position during full inspiration. Average patient anterior–posterior (AP) and left–right (LR) body size was 253.8mm (S.E.=8.3) and 328.3mm (S.E.=6.7) respectively. Images were acquired with a 64mm×0.5mm detector configuration using two CT scanners: Aquilion 64 (n=26) or Aquilion One (n=5) (both Toshiba Medical Systems, Ottawara, Japan). Image acquisition was done with the clinical routine protocol (120kVp, 0.5s rotation time). The raw data from each clinical scan was stored and used for low dose simulation using the Toshiba low dose simulation software.15 No iterative reconstruction techniques were used. Some original data were acquired using automatic dose exposure (n=8, 115–212mA – depending on body habitus), and others with fixed mA settings (50mA). For low dose simulation, the noise levels were reconstructed for the following standard deviation (SD) of noise of 30, 40 and 50 Hounsfield Units (HU) (as allowed by the software). The fixed mA datasets were simulated at 20mA, 15mA and 10mA. Reconstruction was done with various slice thicknesses/increments: 1mm/1mm, 2mm/2mm, 3mm/2.4mm, and 5mm/4mm (for the thicker slice thicknesses an overlap of 20% was chosen to allow for increased characterization benefit16). The reconstruction kernel was FC03 (soft tissue kernel) to reduce noise without limiting imaging characteristics known from CAD. This setting allowed for intra-individual comparison of the impact of dose reduction. As the same raw data set was used, there were no variations in physiological conditions, such as inspiratory depth and fluid content. Only one CT scan was performed on each study participant, thus no additional radiation burden was applied.

Overall, 28 patients with 4 different dose settings and 4 slice thickness/increment settings were reconstructed, in total 448 CT datasets were used for evaluation (Table 1).

The clinical dose CT dataset (original dose, no noise) with 1mm slice thickness was used as reference standard.

All CT images were interpreted in random order and separately by a radiologist (reader 1) with over 15 years of experience in thoracic radiology, and one second year radiology resident (reader 2). Both were blinded to the patients’ clinical history. Lung parenchyma was evaluated by individual lobe (right upper, right middle, right lower, left upper, lingua and left lower) for the presence of GGO, reticulation, and HC. Each lobe was scored for extent of these changes from 0 to 100% in 5% intervals. GGO was defined according to the Fleischner Glossary of Terms for Thoracic Imaging as an increase in lung attenuation without obscuring pulmonary vessels, reticulation as fine or thick reticular grid and thickened interlobular septa and honeycombing as peripheral cysts within a coarse reticulation.17

An overall estimate of image quality, number of lobes involved and total percent of diseased lung was provided by the readers for each patient's series. Image quality was rated on a scale from 1 to 4, with 1 indicating excellent image quality without any artefacts; 2 indicating good image quality with minor streak artefacts without limiting image interpretation; 3 indicating moderate image quality with more severe streak artefacts limiting image interpretation; and 4 indicating bad image quality with severe artefacts significantly impairing image interpretation. The radiologic pattern was recorded as UIP, NSIP, or non-classifiable fibrosis for all CT scans at each LDCT series. UIP and NSIP were defined in accordance with ATS guidelines for IPF and the idiopathic interstitial pneumonias (IIPs) respectively.5,6 As mentioned above, HRCT scans with the highest tube current (clinical established dose levels), 1mm slice thickness were established as the conventional (reference) standard.

The optimal LDCT was selected based on the following three features:(1) diagnostic accuracy, (2) image quality and (3) delivered radiation dose between series. Diagnostic accuracy was considered the most important operating characteristic, given that the overall radiographic pattern often influences clinical diagnosis and treatment decisions. Image quality influences the perception of other CT imaging features (i.e. GGO, reticulation and disease) and hence was chosen as the second most important factor. LDCT imaging series’ with similar diagnostic accuracy and image quality were differentiated based on delivered radiation dose, with the lowest dose series deemed most optimal.

For statistical analysis the average scores for total involvement of GGO, reticulation and HC across all lobes were enumerated. Also, the total disease extent was used (average of the sum of all three types of radiographic changes between the left and right lungs). Mixed linear models were used to investigate the effects of different series on operating characteristics and imaging features. The estimation method “REML” (residual (restricted) maximum likelihood) was used.

Statistical evaluation was done on a per-patient basis using SAS Version 9.3, Cary, North Carolina, USA. A local level of significance (p) of <0.05 was considered statistically significantly.

Of 35 patients approached for study involvement, seven were excluded (4 due to difficulty attaining stored CT images, 2 with hypersensitivity pneumonitis, and 1 with sarcoidosis). Among 28 patients evaluated (56% female) with approximately one-third having a past history of cigarette smoking (Table 2). A clinical diagnosis of IPF and NSIP were made in 20%, and 80% patients respectively. Routine clinical HRCT imaging (120kV, fixed 50mA or dose modulation 115–212mA, 1mm slice thickness) (Table 1), was reported as UIP in 6 cases (24%) and NSIP in 19 (76%) cases. In total, 448 individual CT scans were generated, reviewed and analyzed.

To provide the utmost insight into the data, the evaluation results of reader 1 are provided in Table 3. Given the huge amount of primary data, “Results” section is focused on the statistical results. As mentioned above, for each radiological feature the percentage of affected/diseased lung was visually assessed for each dose level and slice thickness.

The overall analysis found a significant (p<0.001) influence of dose on the correct findings for GGOs (Fig. 1). In detail, for reader 1, the second dose level was not significantly different from the reference standard (p=0.21), while dose level three and four showed significant differences (p=0.0003 and p<0.0005, respectively) (Table 4). For reader 2, even the second dose level led to a significant difference in evaluation of the amount of ggo pattern.

Figure 1.

Presentation of ground appearance using 50mA dose settings simulated 10mA settings (E–H); images are displayed with W/L: 1600/−600HU. 1mm slice thickness (A and E), 2mm slice thickness (B and F), 3mm slice thickness (C and G) and 5mm slice thickness (D and H).

(1.54MB).

The overall analysis found no significant (p=0.25) influence of slice thickness on the correct findings for GGOs for reader 1. In detail, the 2mm slice thickness resulted in a p-value of 1, 3mm and 5mm in p=0.89 and p=0.19, respectively. The combined evaluation of reader 1 and reader 2 confirmed the findings, that a slice thickness of 3mm showed no significant difference (p=0.55).

The overall analysis found a significant (p=0.002 and p<0.0001, respectively) influence of dose on the correct findings for HC. In detail, for reader 1 the second and third dose level were not significantly different (p=0.29 and p=0.11, respectively), while dose level four showed significant difference (p=0.002). For the combined evaluation the second dose level was borderline not significantly different (p=0.55).

The overall analysis for reader 1 found no significant (p=0.30) influence of slice thickness on the correct findings for HC. In detail, the 2mm slice thickness resulted in a p-value of 1, 3mm and 5mm in p=1 and p=0.86, respectively. For reader 2 and the combined analysis a slice thickness of 3mm showed no significant difference on the evaluation of HC, p=0,64 and p=0.68, respectively.

The overall analysis found no significant (p=0.38) influence of dose on the correct findings for reticulation (Fig. 2). In detail, for reader 1 the second dose level was not significantly different (p=0.86), dose level three and four showed p-values of 0.85 and p=0.14, respectively. For reader 2, all reduced dose levels showed a significant difference (p<0.0001).

Figure 2.

Presentation of reticulation using 50mA dose settings (A–D) and simulated 10mA settings (E–H); images are displayed with W/L: 1600/−600HU. 1mm slice thickness (A and E), 2mm slice thickness (B and F), 3mm slice thickness (C and G) and 5mm slice thickness (D and H).

(1.32MB).

The overall analysis (reader 1) found no significant (p=0.59) influence of slice thickness on the correct findings for reticulation. In detail, the 2mm slice thickness resulted in a p-value of 1, 3mm and 5mm in p=0.98 and p=0.94, respectively. For reader 2 the slice thickness had no significant impact on detection rate (mean p=0.14).

The overall analysis found a significant (p<0.0001 and p=0.02, respectively) influence of dose on the correct findings for total disease extend. In detail, for reader 1 the second dose level was not significantly different (p=0.61), dose level three and four showed p-values of <0.0001 and p<0.0001, respectively.

The overall analysis found a significant (p=0.024 and p=0.003, respectively) influence of slice thickness on the correct total disease extend. In detail, for reader 1 the 2mm slice thickness resulted in a p-value of 1, 3mm and 5mm in p=0.98 and p=0.017, respectively.

The overall analysis found a significant (p<0.0001) influence of dose on image quality. In detail, was the second dose level significantly different (p<0.0001), dose level three and four showed p-values of <0.0001 and p<0.0001, respectively.

The overall analysis found a significant (p<0.0001) influence of slice thickness on image quality for both readers. In detail, for reader 1 the 2mm slice thickness resulted in a p-value of =0.08, 3mm and 5mm in p<0.001 and p=0.0002, respectively.

Using a logistic regression model the likelihood for establishing the correct diagnosis was calculated (Fig. 3).

Figure 3.

Diagnostic accuracy based on CT imaging parameters. CT series in milliamperes (mA)/millimetres (mm).

(0.06MB).

The overall analysis found no significant (p<0.36 and p=1, respectively) influence of dose on establishing the diagnosis. In detail, for reader 1 the second dose level was not significantly different (p=1), dose level three and four showed p-values of =0.57 and p=0.12, respectively.

The overall analysis found no significant (p=0.77 and p=1, respectively) influence of slice thickness on establishing the diagnosis. In detail, for reader 1 the 2mm slice thickness resulted in a p-value of =0.6, 3mm and 5mm in p=0.6 and p=0.29, respectively.

The mixed model analysis allows for a combined analysis of dose and slice settings. The estimate of the linear model provides the offset compared to the optimal dataset; the closer this value is to zero, the better.

By using the aforementioned data, it was obvious that a dose level less than 2 leads to significant errors in rating of GGO and total disease extend. Therefore, this specific analysis was tailored to the slice thickness (Table 5). Using this approach, the least estimate (least difference from the original data set) was found for a combination of dose level 2 and slice thickness 5mm for reader 1 and 3mm for reader 2.