Want to keep learning?

This content is taken from the Taipei Medical University's online course, Applications of AI Technology. Join the course to learn more.

Skip to 0 minutes and 14 seconds This is the algorithm the flow chart. Including the ADC map and the DWI image registration. The ADC map was registered to the corresponding DWI by a rigid registration (translation and rotation) and a trilinear interpolation based on the normalized mutual information method to correct for differences due to head movements. The DWI and registered ADC map were normalized so that their intensities were both distributed in a standardized range. From zero to one. The program we used to run this step was the Statistical Parametric Mapping it’s meaning SPM. And the next step is extracting from the brain mask from the whole-brain DWI.

Skip to 1 minute and 28 seconds The brain mask was extracted from the whole-brain DWI based on the estimation of the inner and outer skull surfaces by using BET (Brain Extraction Tool), Step 3 is about preclustering elimination. The histogram of DWI within the brain mask was smoothed by a third-order moving-average filter. The peak of the smoothed histogram was identified. The normalized intensity, denoted by I peak, corresponding to this histographic peak was used as a threshold value. The voxels with normalized intensities lower than or equal to I peak would be eliminated from further processing. Step 4 is about FCM clustering. The remaining voxels after the previous step were divided into 50 clusters by an unsupervised classification with the conventional FCM clustering algorithm.

Skip to 2 minutes and 59 seconds Step 5 is about skimming the clusters from candidate voxels. The clusters with mean normalized intensity larger than the normalized intensity of the histographic peak plus 0.2, that is, I peak plus 0.2, were selected. The voxels belonging to these selected clusters would be treated as candidate voxels of infarct in the next step. Next step is eliminating labels with insufficient intensity. Each cluster of candidate voxels of infarct was further divided into one or several labels, at most the same number as the voxel number in that cluster. A label comprised connected voxels. The labels with average normalized intensity lower than or equal to the threshold value (I peak + 0.2) were eliminated from further processing.

Skip to 4 minutes and 18 seconds The next step is eliminating candidate labels due to magnetic inhomogeneity.

Automated Infarct Detection

In this video, Prof. Peng will introduce the progress of Automated Infarct Detection. You will see the detailed progress of how DWI imaging can identify the cerebral infarct.

Share this video:

This video is from the free online course:

Applications of AI Technology

Taipei Medical University