This paper studies the decentralized output feedback control problem for interconnected stochastic nonlinear systems with dynamic input and output interactions containing time delay. Firstly, we design the decentralized filters independent of time delay to estimate the unmeasured state variables. By estimating the norm of unknown parameter vectors instead of themselves, it can avoid the over-estimation problem. Then, using RBF (radial basis function) neural network to approximate the unknown functions, we construct the adaptive neural network output feedback controller with corresponding adaptive laws. Based on Lyapunov stability theorem, we show that the designed controller can render all the signals of the overall closed-loop systems are semi-global bounded in probability with the help of the changing supply functions. Finally, simulation examples are presented to verify the effectiveness of the theoretic results obtained.