Optimizing sensory feedback via machine learning and in-silico models - amputation results in significant disability, pain, and global societal costs, particularly due to the loss of employ- ment…

The National Institute of Neurological Disorders and Stroke (NINDS) awarded The American University in Cairo a Project Grant of $292,782 under the Extramural Research Programs in the Neurosciences and Neurological Disorders program (CFDA 93.853) on September 8, 2025, with a completion date of August 31, 2027. This award supports the development of a proof-of-concept machine learning-based approach for optimizing sensory feedback in bionic prosthetics. The research integrates U.S. and Egyptian expertise to develop a stimulus coder that uses in-silico computational neuroscience models to predict primary somatosensory cortex activity and identify optimal peripheral nerve electrical stimulation (eStim) parameters. The deliverables include an innovative stimulus coder system designed to enable naturalistic tactile sensations in prosthetic devices, thereby reducing neuropathic pain and improving prosthesis integration and embodiment for amputees.

This research directly addresses a critical public health need in Egypt, where amputation-related disability from labor injuries, road accidents, and diabetes represents significant societal costs and employment loss, particularly among young workers. By advancing prosthetic sensory restoration technologies with limited accessibility in the region, the project aims to enhance the quality of life and functional outcomes for amputee populations while establishing collaborative research capacity between U.S. and Egyptian institutions.

Optimizing sensory feedback via machine learning and in-silico models - amputation results in significant disability, pain, and global societal costs, particularly due to the loss of employ- ment among young workers affected by such injuries. In egypt, this issue is a major public health concern, largely driven by high rates of labor-related injuries (mostly young workers), road accidents, and diabetes. While tradi- tional prostheses can restore some motor control, they do not provide natural sensory feedback or alleviate neuropathic pain, such as phantom limb and residual limb pain, that emerges after injury. Sensory stimulation has been shown to restore sensory function, reduce phantom limb pain, and enhance prosthesis integration, embodiment, and control. However, restoring sensory function in bionic prostheses remains a major challenge. Thus, there is a critical need for advancing prosthetic technologies that restore sensory function, especially in egypt, where access to such innovations is limited. The objective of this project is to develop a proof-of-concept machine learning-based approach for sensory feedback tuning. This project will integrate the computational NEU- roscience expertise of the US PI with the experimental and neural data analysis expertise of the egyptian PI to develop a stimulus coder that optimizes peripheral nerve electrical stimulation (estim) parameters. The design of this stimulus coder is innovative as IT is comprised of an in-silico model that predicts the primary somatosen- sory cortex (S1) activity in response to a given tactile stimulus. The predicted S1 activity is then provided to an estim decoder that identifies the needed estim parameters to evoke the predicted S1 activity. Our central HY- pothesis is that using in-silico computational models to guide the development and operation of the stimulus coder will result in effective naturalistic tactile sensations mimicking those elicited by normal touch. The devel- opment of the proposed stimulus coder will be achieved through two specific aims. In aim 1 of the project, a tactile encoding in-silico model will be developed to simulate primary somatosensory cortex (S1) activity in RE- sponse to tactile stimulation in rats (task US-1). This model will be verified using recorded in vivo RAT tactile data (task EG-1). Next, an estim encoding in-silico model will be developed to simulate S1 activity in response to sciatic nerve estim (task US-2). Similarly, this model will be verified using recorded in vivo RAT estim data (task EG-2). In aim 2, using the estim encoding model of task US-2, expanded datasets will be generated using combinations of different estim parameters, each varied across a wide range, with their evoked neural responses simulated (task US-3). These datasets will be used to train different machine learning techniques for optimal electrical decoding; that is, to identify the optimal estim protocols to generate the desired S1 firing patterns (task EG-3). In addition, the estim strategy most effective for training the decoder will be identified. Successful COM- pletion of this work will result in pilot data that demonstrates the feasibility of developing a smart, adaptable sensory feedback mechanism that could support existing motor prostheses to add sensory capabilities. This is expected to significantly improve the quality of life of amputees in egypt, USA, and worldwide.