Toward Simultaneous Neurostimulation and Prosthetic Control: Real-Time Filtering of Amplitude-Modulated Stimulation Artifacts from Implanted Electrode Signals

Restoring tactile feedback via nerve stimulation is essential for intuitive prosthetic control, yet the electrical artifacts introduced by stimulation pulses-especially with modulating amplitudes-can severely distort electromyographic (EMG) signals and impair decoding accuracy. This challenge is particularly critical in implanted neuromuscular interfaces, where the proximity of stimulating and recording electrodes can increase and magnitude of these signal artifacts. We present a real-time filtering algorithm that adaptively models and subtracts stimulation artifacts from implanted EMG electrode signals by using sample-wise polynomial regression based on stimulation amplitude. The algorithm was implemented on an embedded controller and evaluated in a participant implanted with a long-term neuromusculoskeletal interface compromising an osseointegrated implant for skeletal attachment, intramuscular and epimysial electrodes for prosthetic control enhanced with targeted muscle reinnervation (TMR), and a nerve cuff electrode for sensory feedback. The filter significantly restored key EMG feature distributions in offline tests across multiple movement classes. In real-time motion classification tasks, it improved movement completion rates from 20% to 60% under fixed stimulation and from 32% to 44% under amplitude-modulated stimulation. These findings demonstrate the algorithm's potential for recovering decoding performance during dynamic stimulation and validate its future use in advanced closed-loop prosthetic systems. This work addresses a longstanding barrier to integrating sensory feedback and motor control and provides a foundation for robust, real-world application of bidirectional prostheses using implanted interfaces.