RPS 2210 - Arthritis, inflammation, and sarcopenia

Author Block: K. M. Chapagain, M. Rajeswari Amma, J. Clarke, C. Lowe, T. E. Kirkbride, S. Dahal, S. Gieseg, P. Butler, A. Butler; Christchurch/NZ
Purpose or Learning Objective: The purpose of this study was to evaluate water and lipid component measurement from spectral photon-counting CT for the detection of bone marrow oedema in acute bone injury.
Methods or Background: Patients with acute bone injury were imaged using high-resolution spectral photon-counting CT in the early phase of injury. Physical phantoms were developed to mimic bone marrow and validate water and lipid measurements. The phantoms contained a two-material mixture (water gel, peanut oil) and a three material mixture (water gel, oil and hydroxyapatite nanopowder). Lipid and water maps were generated by harnessing the spectral information contained in the photon-counting CT images. For both phantoms and human images, regions of interest (ROIs) were drawn in the target areas and reference areas to quantitatively measure the water and lipid concentrations. The estimated values from the photon-counting CT were compared with reference values using linearity plots, and the agreement between reference and estimated values were analysed with Bland-Altman plots.
Results or Findings: Estimated water and lipid mass density values had a linear correlation with reference values (linearity=0.98, 0.99) The measurements were not significantly different from reference values (p=0.63, 0.91) with average quantification errors (Bias) (-1.9% and -0.4%), upper limit of agreement (11.5%, 8.7%), and lower limit of agreement (-14.7%, -7.9%) for water and lipid component estimation respectively. Similar to phantom results, the targeted regions in human images showed an increase in water mass density.
Conclusion: Lipid and water components measured from the system are validated using phantom measurements to demonstrate the bone marrow oedema in patients with an acute injury.
Limitations: Comparisons with MRI is not done at this stage, which will be performed in the next phase.
Ethics committee approval: The ethics committee approval was received (18/STH/221).
Funding for this study: This study is funded by MBIE, New Zealand.

Author Block: G. van den Wittenboer¹, A. de Wit¹, E. Langius-Wiffen¹, B. van der Kolk¹, I. M. Nijholt¹, R. van Dijk¹, M. Podlogar¹, M. Maas², M. F. Boomsma¹; ¹Zwolle/NL, ²Amsterdam/NL
Purpose or Learning Objective: To assess the diagnostic accuracy of a cervical spine (C-spine) artificial intelligence (AI) application (Aidoc Medical, Tel Aviv, Israel) for identifying C-spine fractures on CT scans.
Methods or Background: A retrospective diagnostic accuracy study was performed in a level one trauma centre. Consecutive trauma patients (age ≥18 years; 2007-2014) were screened with CT for C-spine fractures. To set the reference standard, one radiologist and three neurosurgeons verified scans considered positive by the radiologist on-call and two radiologists verified negative scans that were flagged positively by the AI application. The index test was defined as detection of ≥1 fracture(s) per scan by the FDA approved and CE marked AI application. The proportion of patients with missed fractures that received stabilising therapy was determined to highlight therapeutic consequences of missed fractures.
Results or Findings: The AI application analysed 2331 patients. After verification, the on-call radiologist’s report was adjusted for 25 patients initially considered negative and flagged positive by the AI (1.2% of all negative scans), increasing the total number of fractures by 13%. The AI application detected 159/211 patients with fractures, resulting in a sensitivity of 75% (95% confidence interval (CI) 69-81%). 16/52 (31%) patients with fractures missed by the AI had received stabilising therapy. Specificity of the AI application was 99% (95% CI 98-99%), overall diagnostic accuracy 97% (95% CI 96-97%), positive predictive value 86% (95% CI 81-90%) and negative predictive value 98% (95% CI 97-98%).
Conclusion: The moderate sensitivity of the AI and the high-miss rate of injuries that received stabilising therapy makes a stand-alone application for screening purposes less expedient, however, as a concurrent reader, AI could aid the radiologist by detecting previously unnoticed fractures, thus increasing the diagnostic yield.
Limitations: Not applicable.
Ethics committee approval: Not applicable.
Funding for this study: Not applicable.