State-of-the-art technology of nano-sensors and actuators enables diagnostics and treatment on the scale of biological cells. However, an advanced medical application should allow the direct feedback of a sensed signal to an actuation, e.g., an action potential propagation through an axon or a special cell activity might be sensed and suppressed by an actuator through voltage stimulation or chemical agent delivery. Such a complex procedure calls for building a communication network between sensors and actuators, but is challenged with issues, such as intensive signal collisions, protocol simplification, and communication efficiency. We propose a collision segregation method and the protocol of collision segregation via learning. Simulations of stress tests show its advantages as compared to the performance of the traditional methods for wireless body area networks. The potential applications are discussed in detail, such as photothermal cancer therapy and anti-cancer drug delivery.