Hello everyone, I am from team 10 for Senior Design from Cal State LA, my name is Christopher Morales, the project is 3D Geometry Reconstruction of Medical Images. Where we're proposing to our clients who are gynecology organs specifically in the bladder, vagina and rectum. Where we will use MRI scans to determine if any patients have a prolapse; which consist of the pelvic floor muscles and ligaments detected being weak and stretched. To identify medical images, we are going to use a method called Computer Vision. Where computer vision is a technique to identify and understand an image. With the use of computer vision, we are able to detect the bladder, rectum, and vagina using MRI Images. Where we will draw an ellipse inside the wanted object in the image to gather information that will be sent to a CAD program. By using the data obtained from the MRI scans through computer vision, we are creating a three-dimensional CAD model of the pelvic floor. In our case, we are utilizing Free CAD because it is an open source parametric 3D modeler which we can feed the information necessary to generate a unique model.
It is important to note that while I have access to MRI images, I cannot share them due to confidentiality agreements with both the CSULA project and the patients. Therefore, as an alternative, I will be utilizing a black and white filtered image of my cat.
Input picture here!!! computer vision example - Arty Chan
The research relevance is based on the notion of 3D Medical Images to grasp an internal representation of manipulating images and give a better observation tool for medical doctors. Implementing Computer Vision to MRI images allows the freedom to interpret MRI images with a 3D render applied, in this case, to the pelvic cavity. Every pelvic organ will be reconstructed with lofted ellipses identified on MRI images that constitutes the 3D volume. Applying a morphological transformation to the MRI images will result in finding the wanted organ to determine the ellipse that contours it to be used for a 3D model.
The 3D Geometry Reconstruction of Medical Images focuses on creating a 3D geometric model that can be then used to achieve numerical simulation aiming at planning surgery. The project will be based on open-source platforms that can use python. Where python allows the project to gather data from a Python IDE and send the data to an open-source CAD to render a model. Each MRI dataset of images will vary on each model output meaning, not every model will be the same.
To get a better understanding of Computer Vision, it is better to understand how an image works despite its looks. Every MRI image consists of three matrices in RGB format that makes a gray image. Where, every element or cell of a matrix will have a value range of 0 to 255, which will be presented as the color of the pixel.
Input picture here!!! MRI open source image - Arty Chan
Furthermore, each MRI image will consist of three matrices that are represented as three channels or known as. Where multiplying the three channels or RBG [1], will result in an image. In computer vision, this is important due to reducing the number of channels and the color range will help the process of elimination of unwanted data.
To create a 3D Pelvic Floor System, the MRI images will go through multiple transformations, elimination, and modifications in order to determine the wanted organ. For team 10, the main goal is to determine a vast number of pixels that can be represented as a blob or will be called a cavity. Where each cavity will have its own correlated ellipse, either the ellipse taking unwanted area or losing area in the blob. Each ellipse will have its own mathematical expression that can be extracted from and sent to FreeCAD software to render and simulate the model.
Input picture here!!! Block diagram - Arty Chan
The requirements for this project do not require any budget. Instead, the approach taken was to use multiple open-source software that are compatible with python. Libraries that are currently running on our software contain pytorch for the deep learning aspect, OpenCV, and Tensorflow. The requirement needed is for the software tool called Miniforge3 which is a conda environment and is able to install python libraries. With Miniforg3, all platforms on any Operating System (OS) will be able to use the python interpreter with the same library version.
One of the necessary implementations for detecting an object from an image is called grayscale. Grayscale is a transformation from RGB, three matrices, that will be converted to one matrix. For the project, reducing as many channels as possible will help detect the wanted image, as the following figure shows.