A bipartite graph approach to retrieve similar 3D models with different resolution and types of cardiomyopathies

Abstract

Three-dimensional (3D) model retrieval uses content-based image retrieval (CBIR) techniques to search for the most similar 3D objects in a dataset, usually considering their geometry and organization in a feature vector. Feature vectors from different objects were compared to establish their similarities. Although this type of comparison typically uses metric distances, such metrics present limitations when the vector lengths are different. Signal-based descriptors are a promising approach for extracting features from 3D objects, but they generate feature vectors with different lengths. Thus, new methods for measuring the similarity are required. This study proposes an approach to 3D model retrieval as a network flow problem using bipartite graphs. The approach was applied to support the diagnosis of cardiomyopathies, considering 3D objects reconstructed from cardiac images of the left ventricle. We achieved an AUC value of 0.93 under the best retrieval scenario. The results also indicate that modeling a 3D model retrieval technique as a network flow problem using graphs can provide a promising manner to compare 3D objects with different shapes and sizes. This strategy, coupled with personal patient data, achieves better results than methods using classical comparison approaches.