Latest CT Technologies in Lung Cancer Screening

Protocols and Radiation Dose Reduction

Marleen Vonder; Monique D. Dorrius; Rozemarijn Vliegenthart


Transl Lung Cancer Res. 2021;10(2):1154-1164. 

In This Article

Future CT Protocols & New Developments

The current guidelines are mainly based on evidence and experience form clinical trials and studies in the past. Continuous efforts are made by the European Imaging Biomarker Alliance (EIBALL) and Quantitative Imaging Biomarker Alliance (QIBA) to streamline the use and validation of imaging biomarkers like lung nodule quantification in lung cancer CT screening. The goal of these subcommittees of the ESR and RSNA is international standardization and harmonization of data acquisition (CT protocol) and analysis (quantification of lung nodules).[30,48] In particular, the QIBA provides some standards for image quality assurance, rather than specified scan parameter values for LDCT. For instance, minimum requirements for image quality defined by the resolution, edge enhancement, HU deviation, voxel noise and spatial wrapping are defined in the 'QIBA Profile: 'Small Lung Nodule Volume Assessment and Monitoring in Low Dose CT Screening'. New (ultra) low dose CT protocols then may not only be tested against the full dose protocol for nodules' sensitivity but could also be benchmarked against these general image quality parameters.

Besides the image quality assurance with the QIBA profile, the EIBALL keeps track of new developments like computer-aided detection (CAD), radiomics and deep learning in the field of lung cancer.[48] The CT protocol can also have impact on the performance of these new applications. In general, use of CAD as a second reader has been found to improve the sensitivity for lung nodule detection.[49] This has also shown to be beneficial in ULDCT. In a study of Takahashi et al. 55 patients underwent ULDCT with a tin-filter protocol. Four thoracic radiologists, with 5 to 21 years of experience, analyzed the low dose scans without and once with the help of CAD as a second-reader for the detection of solid, sub-solid and ground glass nodules of ≥5 mm.[50] The sensitivity could be increased with 19% and 20% for two readers with the help of CAD, whereas for the two others readers the sensitivity was unchanged. In a study by Nomura et al. the performance of CAD software in ULDCT could be improved by using IR algorithms.[51] Training of the CAD software on ULDCT data sets with IR may even further improve the performance of CAD, since the current CAD software was only trained on scans from low-dose protocols.

Another developing field in which the CT protocol plays an important role is radiomics. In the setting of lung cancer screening, numerous other features besides size and shape can be extracted and analyzed from the acquired CT images, that may allow differentiation between benign and malignant nodules.[52] So far, the impact of the CT protocol on radiomic features for nodule characterization has been investigated in a couple of studies. In a phantom and patients study by Lo et al. the effects of dose levels, IR levels and reconstruction kernels on density and texture based features of nodules were analyzed.[53] Density based feature like histogram mean was the most robust feature, while other features were impacted differently by different dose level and reconstruction. The susceptibility of features for protocol changes should be taken into account when density and texture features are used to characterize nodules or nodule change. If disregarded, one might actually measure protocol settings instead of nodule characteristics. Consequently, the authors argue that CT protocol settings should be carefully controlled and only robust features used in quantification and characterization of nodules. Kim et al. investigated the impact of other protocol settings like slice thickness, tube current, and reconstruction kernel using a phantom with spherical nodules.[54] This study showed that all scan parameters significantly affected almost all of the twenty features that were analyzed. Contrary to the former study, the authors of this study recommend standardization and/or normalization of the features that are extracted from scans acquired with different CT protocols. In a patient study by Choe et al. an attempt was made to normalize scans by deep learning image conversion of chest CT scans to improve reproducibility of radiomic features between soft and hard kernels.[55] They showed that the reproducibility of radiomic features could be increased from 15.2% to 57.4% if image conversion was used in data set containing scans with different reconstruction kernels. However, reproducibility of features based on the same kernel scans for two readers was significantly higher with 84.3%. The authors conclude that soft and sharp kernels cannot be used interchangeably and the same reconstruction kernel is required to warrant high reproducibility of radiomic features although deep learning image conversion increased the reproducibility considerably.

New deep learning-based algorithms may not only facilitate standardization based on kernel variation, but also allow to improve the image quality of ULDCT. Research in a phantom has found no significant differences in nodule volume were found based on volume measurements on deep learning post-processed images from ULDCT and the physical nodule volume.[56] Others studies have shown that deep learning denoising techniques can increase image quality compared to FBP and iterative reconstructed ULDCT images and can preserve structural details in the CT image.[57,58] The high image noise present in ULDCT can thus be reduced by these new algorithms, beyond the vendor specific iterative reconstruction techniques, while maintaining image resolution and contrast. This will enable to apply ULDCT in a wider range of screening participants and assists in standardization of CT image quality across CT systems of different vendors.

Currently, the International Association for the Study of Lung Cancer (IASLC) is developing the Early Lung Imaging Confederation (ELIC), with the goal to serve as an global open-source cloud with LDCT images.[59] The ELIC cloud will include high-quality thoracic CT images, that have passed the standardized image quality processes. These high-quality images may then be used in global quantitative lung imaging studies to improve the reliability of clinical decision support in CT lung cancer screening and beyond. The new developments described in this review are very promising to accelerate and optimize the accuracy and precision of lung nodule quantification. After further research, these aspects will have to be included in the standardized protocols and recommendations for lung cancer screening. Recommendations from European, US, and global organizations that include these developments are to be awaited.