Automatic Recognition and Identification of Abdominal Organs in CT Images

Abstract

Recognizing abdominal organs is one of essential steps in visualizing organ structure, for providing assistant in teaching, clinic training, and diagnosis. This paper descirbes a framework for automatic abdominal organ recognition from a series of CT image slices, designed based on shape analysis. Image contextual constraint, and between-slice relationship. Two processing phases, feature extraction (object segmentation) and recognition, are included in this framework. In the phase of object segmentation, a multi-module contextual neural network is applied to segment each image slice into disconnected regions. For each region, its shape features, including relative location, relative distance. Tissue, area, compactness, and elongatedness are calculated, along with its spatial relationships with respect to spine. Futher, according to the knowledge of anatomy, these features are constructed to form fuzzy rules for organ recognition. Followed in the recognition phase, the obtained features and the overlapping information between adjacent slices are used for identifying each organ. This proposed framework has been tested on many clinic patient cases. Results indicate that this framework can successfully recognize abdominal organs, not being affected by partial volume effects.