Automated lesion detection

Fartaria MJ, Bonnier G, Roche A, Kober T, Meuli R, Rotzinger D, Frackowiak R, Schluep M, Du Pasquier R, Thiran JP, Krueger G, Bach Cuadra M, Granziera C. Automated detection of white matter and cortical lesions in early stages of multiple sclerosis. J Magn Reson Imaging. 2015. doi: 10.1002/jmri.25095. [Epub ahead of print]PURPOSE:To develop a method to automatically detect multiple sclerosis (MS) lesions, located both in white matter (WM) and in the cortex, in patients with low disability and early disease stage.
MATERIALS AND METHODS:We developed a lesion detection method, based on the k-nearest neighbour (k-NN) technique, to detect lesions as small as 0.0036 mL. This method uses the image intensity information from up to four different 3D magnetic resonance imaging (MRI) sequences (magnetization-prepared rapid gradient-echo, MPRAGE; magnetization-prepared two inversion-contrast rapid gradient-echo, MP2RAGE; 3D fluid-attenuated inversion recovery, FLAIR; and 3D double-inversion recovery, DIR), acquired on a 3T scanner. To these intensity features we added the information obtained by the spatial coordinates and tissue prior probabilities provided by the International Consortium for Brain Mapping (ICBM). Quantitative assessment was done in 39 early-stage MS patients with a “leave-one-out” cross-validation.
RESULTS: The best lesion detection rate (DR) performance in WM was obtained using MP2RAGE, FLAIR, and DIR intensities (77% for lesions ≥0.0036 mL; 85% for lesions ≥0.005 mL). Similar results were obtained excluding the DIR intensity as well as when using only MPRAGE and FLAIR (DR = 75%, P = 0.5720). However, the combination of FLAIR with DIR and MP2RAGE appeared to be the best for detecting cortical lesions (DR = 62%), compared to the other combination of sequences (P < 0.001).
CONCLUSION: For WM lesion detection, similar results were observed when only conventional clinical sequences (FLAIR, MPRAGE) were used compared to a combination of conventional and “advanced” sequences (MP2RAGE, DIR). Cortical lesion detection increased significantly when “advanced” sequences were used.

DrK was presenting today at the lab meeting about our database of people with MS at Barts. We are also linking up with the MS register. The issue of getting data from scans uploaded was mentioned  tobe a bottleneck because radiographers are snowed under with their day to day MRI duties, so having an automated way to read scans would  be a time saver but would put the imaging centres out of business as the central hubs are used to read scans for clinical trials. This paper reports on attempts to do this, maybe DrK can comment on how accurate it would need to be before we ditch the human touch. 

Sorry I don’t have time to try and explain the different types of images used here I would need  to do some reading.

Maybe we can do it in an educational  post in future. If you click on the tab “education” above their are a few in “teaching posts”

About the author



  • It seems to be a much better subject of interest than brain volume loss. Before implementing a fully automatic lesion detection. What about a little piece of software so that a radiologist (human) could point the center of a lesion by a single click. The radiologist would make as many clicks as there are lesions and the coordinate of these clicks would be recorded. The interest is immediate. Statistics at low cost about where are the lesions, the progression of their number in time + large scale data analysis. This is "simple" to implement, at least much much simpler than a fully automated technique.

  • This might already exist.If yes, are there articles on how the disease is progressing on the anatomical point of view?

By MouseDoctor



Recent Posts

Recent Comments