Image And Mesh Segmentation Of Left Atrial Appendages In Heart CT Images

LAA(left atrial appendage)is an appendage on left atrial.Based on its shape and structure,it is very likely to create thrombus inside for the patients with heart disease.Once the thrombus drops into the left atrial,patients will suffer from a deadly blood vessel embolism.The best treatment on clinical nowadays is conducting a surgery called LAA blocking.The key to the success of the surgery is the good understanding of the LAA condition.In clinic,the judgment of the statue of LAA is totally depending on the experience of the doctor.Because of the heavy burden the doctors take,it is possible that the doctor would make mistake and cause some serious clinic accidents.The doctors are looking for a tool which is smart enough to make the judgment automatically among large number of CT scan data.According to our recent research,we find that there are very few papers in segmenting the LAA.We are trying to search for an algorithm which can locate the LAA.in CT images taken form the patient automatically.After a lot of experiments,we find that there is no suitable image segment algorithm meeting our demand.Therefore we proposed a new image segment algorithm.First,we use Gauss curve fitting algorithm fit the histogram of CT image.Then,we get the threshold from the curve we fit.After thresholding the image with the threshold we gain,we get the left atrial.Further more,we create a mesh model with the surface from the points set of left atrial we gain.We have tried the Ant Colony Optimization(ACO)algorithm and find that it is not suitable for this subject.It is because the quantity of the input vertex in our model is too larger and the topological structure is too complex to fit in for the ACO algorithm.Therefore,we conduct our segmentation with 2 mesh structure parameters,the Shape Diameter Function(SDF)and Geodesic Distance(GD).The SDF makes the surface expand and the GD does the opposite.We keep iterating the process until the two components come to a balance,and the surface will do not change while we conduct more calculation.Then the surface is the LAA result we are hunting for.The algorithm could calculate the result automatically.Compared with the LAA segmentation results provided by our cooperation doctors,we have examined several cases for different patients.The rate of vertex error ranges from-1.4%to 2.1%.The mean error of all model is-0.4%,the SD is 0.031.The algorithm preforming well as expect.And the doctors who support our study a lot confirmed our segment result and speak highly of our work.They confirmed our work is good and could really do a help with them.