Quantitative metal artifact reduction algorithm for spectral CT thermometry.

Spectral CT thermometry can non-invasively monitor internal temperatures to reduce local tumor recurrences caused by insufficient heating/treatment of the tumor and its surrounding safety margin. For its clinical translation, the applied metal artifact reduction algorithm requires quantitative accuracy to ensure the accuracy of generated temperature maps. The newly developed Spectrally Obtained Needle Artifact Reduction (SONAR) algorithm leverages the known shape of the applicator and spectral CT's material decomposition capabilities to isolate the applicator in projections. Projections with long path lengths through metal were then corrected by replacing them with modeled projections of an angled cylinder. To evaluate the accuracy of SONAR, a liver-mimicking phantom embedded with an ablation applicator and thermometers was scanned with a dual-layer spectral CT at phantom temperatures of 35 and 80 °C. Using spectral CT thermometry, temperature maps at 80 °C were generated for image slices with and without the applicator. SONAR significantly decreased streaks along the axis of the applicator. It also eliminated underestimated temperatures immediately adjacent to the applicator and overestimated temperatures in the periphery (2-3 cm from applicator). While application of SONAR resulted in minimal absolute difference in the temperature map without the applicator, averaging 1.1 ± 0.8 °C, temperatures decreased 7.0 ± 4.0 and 10.1 ± 2.3 °C at distances of 2-3 and 0.5-1 cm from the applicator, respectively, to better match the expected temperature. SONAR ultimately minimized metal artifacts and lessened overestimation of temperature in spectral CT thermometry maps. These quantitatively accurate maps will facilitate the evaluation of spectral CT thermometry for non-invasive temperature monitoring of thermal ablations in order to reduce local tumor recurrences.