James Franklin MA, MBBS, Presenter: Nothing to Disclose

Benjamin Irving PhD, Abstract Co-Author: Nothing to Disclose

Margaret Betts MBBS, Abstract Co-Author: Nothing to Disclose

Andre Hallack Miranda Pureza, Abstract Co-Author: Nothing to Disclose

Michael Brady, Abstract Co-Author: Shareholder, Matakina International Limited Shareholder, Mirada Medical Ltd Shareholder, Perspectum Diagnostics Ltd

Julia Schnabel MSc, PhD, Abstract Co-Author: Nothing to Disclose

Fergus Vincent Gleeson MBBS, Abstract Co-Author: Alliance Medical Ltd Consultant

Ewan Mark Anderson MBBCh, Abstract Co-Author: Nothing to Disclose

Pharmacokinetic (PK) modeling of dynamic contrast-enhanced MRI (dceMRI) produces clinically relevant outputs. Accurate tumor delineation is necessary to generate tumor-specific outputs. This study evaluated the impact of interobserver variability in tumor delineation on dceMRI outputs in patients with locally-advanced rectal cancer (LARC).

12 patients with LARC underwent dceMRI at 1.5T before treatment. Two observers delineated tumor volumes on Osirix Medical Imaging Software using the clinical axial small field of view (sFOV) T2W acquisition. Tumor volume delineations were coregistered to the axial T1W dceMRI acquisition using a combined rigid/non-rigid coregistration platform. PK-modeling of the tumor [contrast]-time curve was performed using the Tofts model to derive Ktrans and kep. The two tumor volume delineations were compared using differences in overall volumes and DICE similarity coefficient, which measures the proportion of spatial overlap between two delineations (identical segmentations = 1). Percentage differences in whole-tumor mean, median and variance of Ktrans and kep were calculated. The parameters derived for each observer were compared using paired t-tests and linear regression.  

The mean percentage difference between volumes was 17% (range 1-65%) with mean DICE of 0.77 (range 0.5-0.89). In 9/12 patients DICE was >0.8. The mean percentage differences in mean, median and variance of Ktrans were 3.9% (range 0.5-10.1%), 5.4% (range 0.15-13.4%) and 6% (range 0.5-22%) and equivalent values for kep were 1.7% (range 0.1-5.1%), 2.1% (range 0.4-4.8%) and 4.9% (range 0.4-9.7%). No significant difference was found between the observers (p>0.5) for any of the pharmacokinetic parameters. There was a significant negative correlation between DICE and percentage difference of median (p=0.02) and a similar trend for the percentage difference of mean Ktrans (p=0.17).

In most cases there is good interobserver agreement of rectal tumor delineation. Mean and median of tumour dceMRI pharmacokinetic parameters are relatively robust even for larger discrepancies in delineation, although interobserver variations in Ktrans increase with greater discrepancies in tumor delineation.

There is typically good agreement of rectal tumour volume delineations by trained observers. Mean and median values of dceMRI-derived PK parameters are robust, even for greater disagreement.

Franklin, J, Irving, B, Betts, M, Pureza, A, Brady, M, Schnabel, J, Gleeson, F, Anderson, E, Impact of Interobserver Variability on dceMRI-derived Pharmacokinetic Parameters in Patients with Locally-advanced Rectal Cancer.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14045653.html