Transmission scheduling is a fundamental energy-saving problem in many wireless sensor networks (WSN), especially for mHealth systems with multiple distributed sensing modalities. Recent work has made significant progress to balance the transmission efficiency and timeliness, e.g., periodically entering sleep modes to reduce the power, but such corresponded missing samples impede data integrity and timeliness for realtime diagnosis. In this paper, we intuitively combine transmission scheduling with data imputation, and propose a novel software-hardware cooperated framework for energy-efficient mHealth systems. Specially, we devise a Wasserstein Generative Adversarial Imputation Network (WGAIN) model to impute missing samples, which exploits heterogeneous correlation, temporal dependency, and missing patterns by a divide-and-conquer strategy. We also propose a dropout-based uncertainty approximation method inside the imputation model, prove its equivalence to Gaussian process with variational inference, and explore a heuristic transmission scheduling algorithm for lifecycle management among heterogeneous sensory modules. Extensive experiments on the MIT-BIH dataset and our mHealth prototype have demonstrated the effectiveness compared with state-of-the-art.