For many occasions, segmentation of the bodily tissues are important for diagnosis and surgical planning. However, the segmentation accuracy requirements tend to be more strict for medical images than conventional 2D computer vision problems. The image qualities are often limited by the medical imaging scanners, as well as individual differences.
Several known brain structures, such as the hippocampus , display less distinct borders on most of the MRI images and the individual variabilities may cause less accurate identification when using atlas-registration-based segmentation. Active contour seems to be a good choice for my current interest. However, the level of precision is often questioned. The images can only be effectively obtained in vivo since certain bio-chemical properties will alter on post-mortem samples and thus the image properties for images like SWI; without a proper way of knowing the ground-truth, even the expert-rated result may not help fully verify the quality of the algorithm.
Several known brain structures, such as the hippocampus , display less distinct borders on most of the MRI images and the individual variabilities may cause less accurate identification when using atlas-registration-based segmentation. Active contour seems to be a good choice for my current interest. However, the level of precision is often questioned. The images can only be effectively obtained in vivo since certain bio-chemical properties will alter on post-mortem samples and thus the image properties for images like SWI; without a proper way of knowing the ground-truth, even the expert-rated result may not help fully verify the quality of the algorithm.