Specular reflection removal and bloodless vessel segmentation for 3-D heart model reconstruction from single view images
Three Dimensional (3D) human heart model is attracting attention for its role in medical images for education and clinical purposes. Analysing 2D images to obtain meaningful information requires a certain level of expertise. Moreover, it is time consuming and requires special devices to obtain afore...
No Tags, Be the first to tag this record!
|Three Dimensional (3D) human heart model is attracting attention for its role in medical images for education and clinical purposes. Analysing 2D images to obtain meaningful information requires a certain level of expertise. Moreover, it is time consuming and requires special devices to obtain aforementioned images. In contrary, a 3D model conveys much more information. 3D human heart model reconstruction from medical imaging devices requires several input images, while reconstruction from a single view image is challenging due to the colour property of the heart image, light reflections, and its featureless surface.
Lights and illumination condition of the operating room cause specular reflections on the wet heart surface that result in noises forming of the reconstruction process. Image-based technique is used for the proposed human heart surface reconstruction. It is important the
reflection is eliminated to allow for proper 3D reconstruction and avoid imperfect final output. Specular reflections detection and correction process examine the surface properties. This was implemented as a first step to detect reflections using the standard deviation of RGB colour channel and the maximum value of blue channel to establish colour, devoid of specularities. The result shows the accurate and efficient performance of the specularities removing process with 88.7% similarity with the ground truth.
Realistic 3D heart model reconstruction was developed based on extraction of pixel information from digital images to allow novice surgeons to reduce the time for cardiac surgery training and enhancing their perception of the Operating Theatre (OT). Cardiac medical imaging devices such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT) images, or Echocardiography provide cardiac information. However,these images from medical modalities are not adequate, to precisely simulate the real environment and to be used in the training simulator for cardiac surgery. The propose
method exploits and develops techniques based on analysing real coloured images taken during cardiac surgery in order to obtain meaningful information of the heart anatomical structures.
Another issue is the different human heart surface vessels. The most important vessel region is the bloodless, lack of blood, vessels. Surgeon faces some difficulties in locating the bloodless vessel region during surgery. The thesis suggests a technique of identifying the vessels’ Region of Interest (ROI) to avoid surgical injuries by examining an enhanced
input image. The proposed method locates vessels’ ROI by using Decorrelation Stretch technique. This Decorrelation Stretch can clearly enhance the heart’s surface image.
Through this enhancement, the surgeon become enables effectively identifying the vessels ROI to perform the surgery from textured and coloured surface images. In
addition, after enhancement and segmentation of the vessels ROI, a 3D reconstruction of this ROI takes place and then visualize it over the 3D heart model.
Experiments for each phase in the research framework were qualitatively and quantitatively evaluated. Two hundred and thirteen real human heart images are the dataset collected during cardiac surgery using a digital camera. The experimental results of the proposed methods were compared with manual hand-labelling ground truth data.
The cost reduction of false positive and false negative of specular detection and correction processes of the proposed method was less than 24% compared to other
methods. In addition, the efficient results of Root Mean Square Error (RMSE) to measure the correctness of the z-axis values to reconstruction of the 3D model accurately
compared to other method. Finally, the 94.42% accuracy rate of the proposed vessels segmentation method using RGB colour space achieved is comparable to other colour
spaces. Experimental results show that there is significant efficiency and robustness compared to existing state of the art methods.