Neuroprosthetic devices that use transcutaneous neuromuscular electrical stimulation (NMES) are potential interventions to restore skeletal muscle function in people with neurological disorders. As commonly noted, how to assess the NMES-induced muscle fatigue is a critical problem. This is because the capability of fatigue assessment is a necessary precursor for optimally modulating the NMES dosage to improve the control performance of a neuroprosthesis and ensure user’s safety. To effectively estimate the NMES-induced muscle fatigue, this paper proposes a novel state observer that combines a mathematical predictive fatigue model and intermittent feedback from ultrasound-derived strain images. The strain images quantify muscle contractility during NMES. Principal component regression (PCR) is used to derive a relationship between the strain images and instantaneous muscle force production. Lyapunov stability analysis was performed to obtain the convergence property of the designed observer. A globally uniformly ultimately bounded (GUUB) result was obtained. Simulations based on pre-recorded data from a human experiment were also conducted to demonstrate the performance of the designed observer.