Evaluation of 4 three-dimensional representation algorithms in capsule endoscopy images

Area of Science:

• Medical Imaging
• Computational Geometry
• Gastroenterology

Background:

• Small-bowel capsule endoscopy (SBCE) generates 2-D images of the gastrointestinal tract.
• Accurate 3-D reconstruction of SBCE images is crucial for diagnosing abnormalities.
• Publicly available Shape-from-Shading (SfS) algorithms offer potential for 3-D reconstruction.

Purpose of the Study:

• To evaluate and compare the performance of four publicly available Shape-from-Shading (SfS) algorithms for 3-D representation of SBCE images.
• To identify the most effective SfS algorithm for clinical SBCE applications.

Main Methods:

• Four SfS algorithms were applied to 54 2-D SBCE images (protrusion, inflammation, vascular categories).
• Three expert reviewers assessed the 3-D reconstructions generated by each algorithm.
• Inter-rater agreement was calculated to ensure reliability of the evaluations.

Main Results:

• Tsai's SfS algorithm demonstrated superior performance, being selected as the best in 45 out of 54 images.
• Tsai's algorithm was unanimously chosen as the best in 26 images, irrespective of lesion type or capsule system.
• Good inter-observer agreement (kappa = 0.55) was achieved, indicating consistent results.

Conclusions:

• 3-D representation software using SfS algorithms, particularly Tsai's, offers a practical alternative for visualizing SBCE findings.
• This approach provides a valuable tool for 3-D reconstruction until advanced hardware enables true 3-D imaging in SBCE.