Adaptive Localizing Region-Based Level Set for Segmentation of Maxillary Sinus Based on Convolutional Neural Networks

Figure 1

Inhomogeneous appearance of maxillary sinus shows challenges in research of segmentation. (a) A pair of healthy cavity and (b–d) lesions, sizes, locations, and shapes which are diffused and reveal the wide spatial distribution. Failed segmentation is more likely caused by incorrect noisy information. (e) The average of the normalized intensity histograms of corresponding CT images in maxillary sinus. Different distributions can be observed easily and demonstrate difficulty for most popular methods.