New algorithm using an adaptive level set model applied to hippocampus segmentation and volume calculation in MRI images

The hippocampus is known as one of the most important brain structures since the change in its volume is an early symptom of many diseases such as Alzheimer's disease. Accurate measurement of the hippocampus is very helpful in identifying lesions. Hippocampus segmentation in MRI images is of vital importance for the in-depth study of many brain diseases. However, hippocampus segmentation remains a difficult task due to its small size, complex shape, as well as its imaging characteristics with low contrast and weak and blurred boundaries. To overcome these problems, in this paper, an efficient process is suggested by modeling and solving a system of two-dimensional partial differential equations (PDEs). The first equation allows for restoration using Euler's equation, similar to anisotropic smoothing based on a regularized Perona and Malik filter that removes noise while preserving edge information. The second equation uses the adaptive level set method to segment the image based on the solution of the first equation. This process takes place alternately between these two equations until convergence. This approach allows developing a new algorithm that overcomes the studied model drawbacks. Results of the proposed method give clear segments that can be applied to any application. The proposed method is applied for the hippocampus volume calculation associated with the Scheltens scale. Performance evaluations compared automatic segmentations with manual segmentations performed by expert radiologists. The results revealed a Dice similarity rate average 91.8% and with a volume value varies between 3.19 cm3 and 1.3 cm3 for the right hippocampus in the Scheltens scale, which represents a very significant value compared with other work in the field. Therefore, the method proves clinically useful and effectively segments the hippocampus and gives credibility to the volume calculation, which shows that the developed approach produces superior results in terms of quantity and quality compared to other models already presented in previous works.