RSNA 2016

Abstract Archives of the RSNA, 2016


Image Processing to Improve Visualization of Mesial Temporal Sclerosis on MRI

Tuesday, Nov. 29 3:00PM - 3:10PM Room: N229

Matthew S. Parsons, MD, Saint Louis, MO (Abstract Co-Author) Nothing to Disclose
Aseem Sharma, MBBS, Saint Louis, MO (Presenter) Stockholder, General Electric Company Consultant, BioMedical Systems Co-Founder, Correlative Enhancement, LLC
Rishi Mhapsekar, MD, Saint Louis, MO (Abstract Co-Author) Nothing to Disclose
Charles F. Hildebolt, DDS, PhD, Saint Louis, MO (Abstract Co-Author) Nothing to Disclose

Mesial Temporal Sclerosis (MTS) may be difficult to detect on MRI if associated volume loss and signal alteration are minimal. Aim of this study was to test whether an image-processing algorithm can aid in detection of MTS by increasing the contrast between abnormal hippocampus and normal brain.


We selected coronal FLAIR Images across hippocampi of 18 adult patients (10 females, 8 males; mean age 41.2 years) with proven MTS. One investigator who did not participate in patient selection, image analysis or statistical analysis processed these images in MATLAB using an algorithm (patent pending, Correlative Enhancement LLC) that aimed to selectively increase intensity of abnormal hippocampus. Another investigator recorded signal intensity of hippocampi, gray matter, and white matter for baseline and enhanced images placing equivalent ROIs. Standard deviation in air signal was used as a measure of noise. Using these measurements, we calculate average (Hmean) and maximum (Hmax) contrast-to-noise ratio (CNR) between abnormal hippocampus and white matter. We also calculated CNR for normal extralimbic gray matter (GM). In separate sessions, a neuroradiologist rated signal intensity of hippocampi on baseline and enhanced images on a 5-point scale ranging from 1(definitely abnormal) to 5 (definitely normal). Differences in Hmean, Hmax, GM, and SI ratings for baseline and enhanced images were assessed for statistical significance.


At baseline, both Hmean (mean 30.0, 95% CI 24.5-35.5) and Hmax (mean 33.5, 95% CI 27.3-39.6) were higher than GM (mean 20.5, 95% CI 15.0 -24.9, p<0.0001). Image processing resulted in significant increase in Hmean (mean 40.4, 95% CI 33.3-47.5; p < 0.0001) and Hmax (mean 50.9, 95% CI 41.6 – 60.3; p < 0.0001) without affecting GM (mean 20.5, 95% CI 16.3-24.7; p = 0.9375). SI ratings showed a more confident identification of abnormality on enhanced images (p=0.0001).


In an experimental setting, image-processing algorithm resulted in selective improvement in CNR of abnormal hippocampus with associated improved visual conspicuity of hippocampal signal alteration.


It may be feasible to improve MRI detection of MTS using image-processing algorithms that selectively enhance CNR of abnormal hippocampus.