Cube Kohonen self-organizing map (CKSOM) model with new equations in organizing unstructured data.

Area of Science:

• Computer Vision
• Data Science
• Medical Imaging

Background:

• Surface reconstruction from 3-D data is crucial for object representation.
• Unstructured 3-D data lacks point connectivity, leading to inaccurate shape representation.
• Existing Kohonen self-organizing maps (KSOM) have limitations with closed 3-D surfaces.

Purpose of the Study:

• To adapt and enhance the Kohonen self-organizing map (KSOM) for organizing unstructured 3-D data.
• To improve surface reconstruction accuracy for closed surfaces using KSOM.
• To evaluate KSOM's performance on medical imaging data.

Main Methods:

• Developed an enhanced 3-D KSOM model with added class number and index vector.
• Introduced new equations for data organization and topology correction.
• Tested various grid sizes and maximum iterations, implementing the model in Dev C++ and using GNUPlot for visualization.

Main Results:

• The number of data redundancies is directly proportional to the grid size used in KSOM.
• Increasing maximum iterations results in a smoother reconstructed surface.
• The enhanced KSOM effectively organizes unstructured data for closed surfaces, validated by an area formula.

Conclusions:

• The enhanced KSOM model successfully organizes unstructured 3-D data for accurate surface reconstruction of closed objects.
• The study demonstrates KSOM's applicability to medical imaging, overcoming limitations of previous 2-D and basic 3-D approaches.
• The findings provide a more robust method for handling complex 3-D data in various scientific fields.