Systematization of computed tomography reconstruction filters for artificial intelligence algorithms using the example of chest and brain organs: a retrospective study

The selected convolution kernel in computed tomography (CT) directly affects the results of artificial intelligence (AI) algorithms. The formation of uniform requirements for this parameter is complicated by the fact that such filters are unique to equipment developers. The aim of the work is to create a table of correspondence of reconstruction filters between different equipment manufacturers to direct to the AI algorithms the series of images on which, in CT of the chest organs and the brain, the quantitative analysis will be most reproducible. DICOM tags 0018,1210 (Convolution Kernel), 0008,0070 (Manufacturer), 0018,0050 (Slice Thickness) of CT images from the Unified Radiology Information Service of Moscow were downloaded and analyzed. Inclusion criteria: age older than 18 years; slice thickness ≤ 3 mm. The data analysis is presented in the form of summary tables comparing reconstruction filters from different manufacturers for chest and brain CT, a number of clinical tasks, as well as descriptive statistics of their distribution by scanning area and manufacturer. 1905 chest ("CHEST" and "LUNG") and brain ("HEAD", "BRAIN") CT studies were included in the analysis. In chest CT, reconstructions to evaluate pulmonary parenchyma and mediastinal structures were common. Reconstructions for brain parenchyma and bone structures were common in brain CT. Systematization of reconstruction filters for chest and brain CT was performed. The obtained data will allow correct image series selection for quantitative AI analysis.

1. Vasilev Yu.A., Vladzymyrskyy A.V., Bondarchuk D.V., Kozhikhina D.D., Reshetnikov R.V., Blokhin I.A., Solovev A.V., Gatin D.V. Importance of artificial intelligence technologies to prevent defects in radiologist's practice. Medical Doctor and Information Technology. 2023;(2):16–27. (In Russ.). https://doi.org/10.25881/18110193_2023_2_16

2. Schaller S., Wildberger J.E., Raupach R., Niethammer M., Klingenbeck-Regn K., Flohr T. Spatial domain filtering for fast modification of the tradeoff between image sharpness and pixel noise in computed tomography. IEEE Transactions on Medical Imaging. 2003;22(7):846–853. https://doi.org/10.1109/tmi.2003.815073

3. Blokhin I.A., Rumyantsev D.А., Suchilova M.M., Gonchar A.P., Omelyanskaya O.V. Low-dose computed tomography in COVID-19: systematic review. Digital Diagnostics. 2023;4(1):25–37. https://doi.org/10.17816/DD119870

4. Boedeker K.L., McNitt-Gray M.F., Rogers S.R., Truong D.A., Brown M.S., Gjertson D.W., Goldin J.G. Emphysema: Effect of Reconstruction Algorithm on CT Imaging Measures. Radiology. 2004;232(1):295–301. https://doi.org/10.1148/radiol.2321030383

5. Wang Y., De Bock G.H., Van Klaveren R.J., Van Ooyen P., Tukker W., Zhao Y., et al. Volumetric measurement of pulmonary nodules at low-dose chest CT: effect of reconstruction setting on measurement variability. European Radiology. 2010;20(5):1180–1187. https://doi.org/10.1007/s00330-009-1634-9

6. Behrendt F.F., Das M., Mahnken A.H., Kraus T., Bakai A., Stanzel S., et al. Computer-Aided Measurements of Pulmonary Emphysema in Chest Multidetector-Row Spiral Computed Tomography: Effect of Image Reconstruction Parameters. Journal of Computer Assisted Tomography. 2008;32(6):899–904. https://doi.org/10.1097/rct.0b013e31815ade64

7. Vasil'ev Yu.A., et al. Komp'yuternoe zrenie v luchevoi diagnostike: pervyi etap Moskovskogo eksperimenta. Izdatel'skie resheniya; 2023. 376 p. (In Russ.).

8. Mackin D., Ger R., Gay S., Dodge C., Zhang L., Yang J., et al. Matching and Homogenizing Convolution Kernels for Quantitative Studies in Computed Tomography. Investigative Radiology. 2019;54(5):288–295. https://doi.org/10.1097/rli.0000000000000540

9. Dalrymple N.C., Prasad S.R., El-Merhi F.M., Chintapalli K.N. Price of Isotropy in Multidetector CT. RadioGraphics. 2007;27(1):49–62. https://doi.org/10.1148/rg.271065037

10. Kauermann G., Küchenhoff H. Stichproben: Methoden und praktische Umsetzung mit R. Heidelberg: Springer Berlin; 2010. 261 p. https://doi.org/10.1007/978-3-642-12318-4

11. Blokhin I.A., Kodenko M.R., Shumskaya Y.F., Gonchar A.P., Reshetnikov R.V. Hypothesis testing using R. Digital Diagnostics. 2023;4(2):238–247. https://doi.org/10.17816/dd121368

12. Fu B., Wang G., Wu M., Li W., Zheng Y., Chu Z., et al. Influence of CT effective dose and convolution kernel on the detection of pulmonary nodules in different artificial intelligence software systems: A phantom study. European Journal of Radiology. 2020;126. https://doi.org/10.1016/j.ejrad.2020.108928

13. Hoang-Thi T.-N., Vakalopoulou M., Christodoulidis S., Paragios N., Revel M.P., Chassagnon G. Deep learning for lung disease segmentation on CT: Which reconstruction kernel should be used? Diagnostic and Interventional Imaging. 2021;102(11):691–695. https://doi.org/10.1016/j.diii.2021.10.001

14. Lu L., Ehmke R.C., Schwartz L.H., Zhao B. Assessing Agreement between Radiomic Features Computed for Multiple CT Imaging Settings. PLoS ONE. 2016;11(12). https://doi.org/10.1371/journal.pone.0166550

15. Berenguer R., Pastor-Juan M. del R., Canales-Vázquez J., Castro-García M., Villas M.V., Mansilla Legorburo F., Sabater S. Radiomics of CT Features May Be Nonreproducible and Redundant: Influence of CT Acquisition Parameters. Radiology. 2018;288(2):407–415. https://doi.org/10.1148/radiol.2018172361