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

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 x 10 x 28 cm(3)) was filled with heated water (35.0-38.8 degrees C) and used as a water phantom. The DWI data of 23 healthy subjects (aged 26-75years; mean +/- standard deviation, 50.13 +/- 19.1years) were used for this study. The temperature was calculated using the following equation: T(degrees 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 degrees 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 degrees 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 degrees 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 degrees C, respectively). In humans, however, the histogram curve-fitting methods yielded temperatures approximately 1 degrees C higher (+1.04 degrees C and +1.36 degrees 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. Copyright (C) 2011 John Wiley & Sons, Ltd.