A data mining approach to the SAR values over large MR image repositories

Abstract

Purpose: In magnetic resonance imaging, the radiofrequency energy absorption arises as one of the main safety concerns, being mainly related with increased body temperature. Monitoring radiofrequency absorption is achieved by the estimation of specific absorption rate (SAR), whose implementation lies on equipment manufacturers, which in turn are not totally enlightening about its calculus. This work presents an exploratory approach of whole-body SAR values stored in DICOM metadata aiming to find correlation with body weight, body mass index (BMI), gender and pulse sequences for abdominal/pelvic (17.812 series) and head (29.907 series) studies. Methods and Materials: All studies were acquired in a 3 Tesla scanner with high-performance gradients. Data were extracted using Dicoogle, a DICOM metadata mining tool. Several DICOM tags were analysed (e.g. patient weight, height, gender, sequence name). For each study type, specifically weighted pulse sequences were related with weight, BMI and gender through boxplot diagrams, statistical and effect size analysis. Results: SAR limits were never exceeded. Generally, SAR values tended to decrease with increasing body weight and BMI values for abdominal/pelvic studies. On the other hand, head studies showed different trends regarding distinct pulse sequences. SAR values tend to be higher in male individuals (p<0,05). As expected, turbo spin echo sequences present the highest SAR values. The values found for echo gradient spoiled sequence (FLASH) were also high. Conclusion: It is confirmed that SAR estimates are related with the analysed variables. An individual examination of pulse sequences is recommended to observe trends regarding weight, BMI or gender.