Guided review

The idea of guided review is to use the extracted spatial distribution patterns to guide a reader to specific segments that were not labeled as having plaques but that show a high probability of containing initially missed plaques. So, guided review is intended to support the detection of initially missed plaques and hence to increase the robustness of plaque detection.

For this, a set of prediction rules has to be learned in advance from an appropriate population. Afterwards, these prediction rules can be used to search for initially missed plaques. To this end, the current plaque labeling of a patient is taken and the items are computed according to the defined itemset. These items are then regarded as possible conditions and the set of prediction rules is searched if - with the given condition items - prediction rules exist with a segment as conclusion that has so far not been labeled as having plaques.

In the study "Guided review by frequent itemset mining: additional evidence for plaque detection" we have shown that guided review significantly performs better than a weighted random approach for detection of initially missed plaques.

Furthermore, we have developed a web application were you can try guided review on your own.