Calculation methods for ventricular diffusion-weighted imaging thermometry: phantom and volunteer studies

Abstract

A method for the measurement of temperature in the lateral ventricle using diffusion-weighted imaging (DWI) has been proposed recently. This method uses predetermined arbitrary thresholds, but a more objective method of calculation would be useful. We therefore compared four different calculation methods, two of which were newly created and did not require predetermined thresholds. A rectangular polyethylene terephthalate bottle (8 × 10 × 28 cm(3)) was filled with heated water (35.0-38.8 °C) and used as a water phantom. The DWI data of 23 healthy subjects (aged 26-75 years; mean ± standard deviation, 50.13 ± 19.1 years) were used for this study. The temperature was calculated using the following equation: T(°C) = 2256.74/ln(4.39221/D) - 273.15, where D is the diffusion coefficient. The mean ventricular temperature was calculated by four methods: two thresholding methods and two histogram curve-fitting methods. As a reference, we used the temperature measured at the tympanic membrane, which is known to be approximately 1 °C lower than the brain temperature. The averaged differences in temperature between mercury thermometry and classical predetermined thresholding methods for the water phantom were 0.10 ± 0.42 and 0.05 ± 0.41 °C, respectively. The histogram curve-fitting methods, however, yielded temperatures a little lower (averaged differences of -0.24 ± 0.32 and -0.14 ± 0.31 °C, respectively) than mercury thermometry. There was very little difference in temperature between tympanic thermometry and classical predetermined thresholding methods (+0.01 and -0.07 °C, respectively). In humans, however, the histogram curve-fitting methods yielded temperatures approximately 1 °C higher (+1.04 °C and +1.36 °C, respectively), suggesting that temperatures measured in this way more closely approximate the true brain temperature. The histogram curve-fitting methods were more objective and better matched the estimated brain temperature than did classical predetermined thresholding methods, although the standard deviation was wider in the former methods.