A neural network model of phantom limbs.

This paper presents a detailed clinical description of phantom limbs and a neuronal network model that provides a comprehensive and parsimonious explanation of otherwise inexplicable or at least unrelated phenomena. Simulations of self-organizing feature maps (Kohonen networks) that had been trained to recognize input patterns were deprived of parts of their input in order to simulate partial deafferentation. This leads to reorganization processes that are shown to be driven by input noise. In patients with an amputated limb, this noise is generated by dorsal root ganglion sensory neurons which are known to fire irregularly upon laceration. According to this model, the long-standing debate concerning non-cortical vs. cortical contributions to the generation of the phenomenon of phantom limbs can be resolved in that it is the peripherally generated noise that causes cortical reorganization. The model can be tested and may have therapeutic implications.