Automated angiodysplasia detection from wireless capsule endoscopy.

We present a novel system for the automatic detection of angiodysplasia lesions from capsule endoscopy images. The approach identifies potential regions of interest and classifies them using a combination of color-based, texture, statistical and morphological features. A boosted decision tree classification method is used in order to overcome the problem of unbalanced sampling between pathological and non-pathological regions. The lesion detection method has been designed and validated using a lesion database labelled by an expert. The approach achieves a sensitivity of 89.51% and a specificity of 96.8%, thus providing a high performance in the detection of angiodysplasia lesions.