Automatic hand phantom map generation and detection using decomposition support vector machines.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Human-Computer Interaction

Background:

• Sensory feedback for myoelectric prosthesis users can improve object manipulation and embodiment.
• Phantom maps on residual limbs offer a target for non-invasive tactile feedback.
• Accurate detection of phantom digit boundaries is challenging due to individual variations.

Purpose of the Study:

• To develop and evaluate automatic methods for detecting phantom map boundaries.
• To compare the performance of different support vector machine (SVM) algorithms and sampling strategies.
• To assess the impact of stimulation array density on detection accuracy.

Main Methods:

• Proposed automatic phantom map detection using four SVM decomposition algorithms and three sampling methods.
• Incorporated fuzzy logic and active learning strategies.
• Tested algorithms on generated and reported phantom map datasets using dense (100x100) and coarse (3x5, 4x6) stimulation arrays.

Main Results:

• Majority-pooling sampling yielded the smallest error rates.
• One-vs-one SVM architecture showed higher classification accuracy.
• Fuzzy logic reduced noise, and active learning improved accuracy.
• Dense arrays with one-vs-one fuzzy SVM and majority-pooling achieved the lowest average error (8.78% - 11.5%).

Conclusions:

• SVMs with dense arrays effectively detect refined phantom map shapes.
• Coarse arrays are unsuitable for precise detection.
• Proposed a two-step approach using a dense array for detection and a customized coarse array for stimulation.
• Methodology aids haptic feedback design and phantom map tracking.