Segmentation in CT scan and multimodal MRI datasets using 3D active contours

Context: Segmentation is an important tool in the diagnosis and prognosis as well as for the treatment planning of medical disorders. In hospitals, the two main departments that need segmentation are the neurosurgery and the radiotherapy departments. The first one uses currently mainly manual segmentation tools while the second one uses also AI-based and atlas-based segmentation the easiest applications. Objective: The objective of this work is to develop and implement a user-friendly semi-automatic tool to perform segmentation on medical images that would reduce the task time taken by the doctors. Method: The method developed uses a region-based 3D active contour algorithm using robust statistics which allows good interactions with the user. To support this algorithm, a graphical user interface was implemented to enable users to easily used this segmentation tool. This method was tested in two applications; the segmentation of femoral heads in CT scans and brain pathologies in MRI. Results: The results show that the algorithm produces promising results with mean dice coefficients of 0.868 and 0.9016 for the two applications. This algorithm helps dividing the drawing time by a mean factor of 6.996 compared to software used in hospitals. Conclusion: This region-based active contour algorithm is an interesting and flexible segmentation tool. It allows to have good interactions with the user and, with a good GUI, can be used by doctors in their daily work for a variety of applications. The tool could also be inserted in the workflow of an active learning segmentation scheme to further improve the segmentation process and provide the medical staff with a new efficient and robust way to label images.