John Hasop Woo MD, Presenter: Nothing to Disclose

L. Pikus, Abstract Co-Author: Nothing to Disclose

Jaroslaw Krejza MD, Abstract Co-Author: Nothing to Disclose

Elias Rafic Melhem MD, Abstract Co-Author: Nothing to Disclose

Observer performance in the detection of MS plaques is difficult to measure using real MR images due to confounding biases. We conducted a free-response receiver operating characteristic (FROC) study to assess performance using phantom MR images with simulated MS lesions.

MR parametric maps were derived at supra- and infratentorial levels from a mixed multi-echo inversion recovery scan of a healthy volunteer. Similar maps were obtained of a lesion in a patient with MS. Pixel-wise solution of the Bloch equation enabled us to create images of either T2-weighted (T2W) (TR/TE: 4500/100ms) or FLAIR (TR/TE/TI: 11000/140/2600ms) contrast, in which we could place simulated lesions of varying size, number, and location. In 160 images, 80 at either axial level, a total of 240 lesions were placed pseudo-randomly in the brain. Automated software displayed the same 160 images to four neuroradiologists of comparable training, who rated suspicious lesions on a 4-point scale reflecting their confidence level. Undetected “true” lesions were assigned a rating of 0. Observer performance was measured by the area under the alternative FROC curve (A1), as estimated by the JAFROC algorithm.

Pooled A1 scores were significantly better for FLAIR than T2W images supratentorially (0.96±0.02, 0.83±0.04; P<0.01, Z-test), but were similar infratentorially (0.86±0.08, 0.83±0.09). Scores also varied with lesion location: for FLAIR, mean A1 values for cortical, deep white matter, and periventricular lesions were 0.92±0.07, 0.98±0.01, and 0.84±0.06, respectively; for T2W, they were 0.63±0.11, 0.96±0.04, and 0.88±0.08. Among the locations, FLAIR scores were significantly higher than T2W only for cortical lesions (P<0.03). We also found a linear correlation between A1 and lesion size (r=0.5, P<0.05).

By our measures, observer performance in the detection of supratentorial MS lesions on FLAIR images is excellent and better than T2W images, while infratentorial performance is moderate and similar on both modalities. Performance on FLAIR images also exceeds T2W in detection of cortical lesions. To generalize our results to the general population, however, a larger number of observers needs to be tested.

Woo, J, Pikus, L, Krejza, J, Melhem, E, Observer Performance in Detection of Simulated Multiple Sclerosis (MS) Lesions on T2-weighted and FLAIR Images of the Brain.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4414789.html