This research involves the use of 2-D images obtained from an ingestible MicroCam capsule to make a 3-dimensional depiction of the human gastrointestinal (GI) tract. With the assistance of an external Rotating Permanent Magnet, the capsule will be guided through the tract whilst recording image data. Using an image depth estimation model called Depth Anything V2, an approximate distance between the camera and objects within the images can be made.
The results are represented by pixels forming a collection of data points, each possessing a set of coordinates on the X, Y and Z axes. This is known
as a Point Cloud. Navigation data of the capsule might also be incorporated to accurately tell the relativity between the points so as to form a continuous image. These individual points are then combined to form a 3-dimensional mesh-like structure, shaping the surface of the object depicted. A helpful algorithm that can assist in achieving this is a Marching Cube Algorithm. The finished product will be the 3-dimensional reconstruction of the images. A phantom of the GI tract will be used in navigation experiments for the capsule. Using a pre-existing 3D model of this phantom (for example, what it was based on), the accuracy of the reconstruction can be determined by comparing the dimensions of both sources.
This is in the hopes of making the detection of lesions easier, which in turn improves the diagnostic accuracy of endoscopic procedures.