Purpose: To investigate the effects of head posture on dosimetry of non-contrast head CT scans, facilitated by a deep learning (DL) based method.
Methods: Non-contrast head CT exams of 6 months were collected and all exams were acquired on a 2nd-generation dual source CT scanner (Flash, Siemens) with identical clinical protocol: 100kV, 35mA for topogram; 120kV, 350reference mAs, pitch0.55, and 128x0.6 mm for acquisition. A U-Net based DL method was adopted and trained to calculate the angle between orbitomeatal line (OML) and anterior-posterior (AP) direction from topogram images of each patient. Outer canthus of eye and external auditory meatus were manually labeled as landmarks of OML for 136 exams. Labeled exams were randomly split into training (60%), validation (20%), and evaluation (20%). The trained DL was then used to calculate OML-AP angle for the rest 2193 exams. Linear regression was applied to test the relation between CT dose metrics (CTDI, DLP), scan length and the absolute values of angles. A head phantom was placed at 3 different postures and scanned with clinical head protocol, respectively.
Results: The RMSE between true and predicted OML-AP angle is 1.8°, indicating reliable head posture measurements. The OML-AP absolute angles of 2193 exams range from 0° (perfect axial with appropriate head tilt) to 43.3° (no head tilt), with a median value of 7.2°. DLP, CTDI and scan length are positively correlated with OML-AP angle with slope of 4.3mGy•cm/degree, 0.11mGy/degree and 0.05 cm/degree, respectively. Phantom results suggest similar trend that DLP increases from 516.2 to 582.5mGy•cm, CTDI slightly increases from 41.0 to 41.3mGy, and scan length increases from 12.6 to 14.1cm, when OML-AP angle increases from 3.7° to 31.9°.
Conclusion: This study demonstrated the importance of head tilting in clinical CT exams that proper posture can reduce >13% DLP during head CT exams.
Not Applicable / None Entered.