Fueled by healthcare electronics (HE) and artificial intelligence (AI), smart healthcare market has been emerging, such as medical wireless body area network (WBAN), which can be used for daily health and early disease monitoring, where intra-WBAN communication is at the first tier. However, timely message sending and low-power consumption are still two essential issues. Therefore, an AI-generated content-based control strategy called KsCS-DQNs is suggested to address these challenges. Firstly, we establish an asymmetric two-level polling control model for intra-WBAN communication. Then, in order to balance the performance metrics of mean waiting time (MWT), mean queue length (MQL) and fairnes, an optimization algorithm is proposed to handle the optimization problem and achieve convergence in dispersed action space. By utilizing multi-DQN to generate access order and number of health packets that are sent for each node, the optimal MWT and MQL under fairness are gained. Additionally, connection and beacon frames and superframe are simplified to reduce the control overhead, which contributes to the encapsulation of nodes with limited energy and improves their accessibility. Experiments show that KsCS-DQNs outperforms classical methods, such as IEEE 802.15.6 PCF, DQN-P and two-level Gated polling control, especially in an asymmetric system with randomly emergent traffic.