Abstract: Spectral computed tomography (CT) enhances material decomposition capability and clinical diagnostic accuracy by acquiring projection data from multiple energy channels. However, under low-dose scanning conditions, insufficient photon counting leads to severe quantum noise and streak artifacts in reconstructed images. The directional probabilistic total variation (dTV-p) regularization demonstrates excellent performance in multi-channel reconstruction by dynamically selecting reference channels via a probability mass function. However, its reliance on intermediate iterative results to construct reference images often introduces noise and artifacts, which limits the final reconstruction quality. To address this limitation, in this paper, we have proposed a denoising convolutional neural network (DnCNN)-like oriented probabilistic total variation algorithm for spectral CT reconstruction. For the first time, we integrated a pre-trained residual learning-based DnCNN-like denoising network into the dTV-p regularization framework. During each iteration, the reference image, dynamically selected according to the probability mass function, is intelligently enhanced to generate cleaner and more accurate directional gradient guidance. Specifically, the algorithm first constructs a probability distribution based on the geometric mean of the signal-to-noise ratios across channels to select the reference channel. Then, a deep residual network pre-trained on large-scale CT image datasets is used to denoise and structurally enhance the reference image. Finally, in combination with a weighted least squares data fidelity term, an overall optimization model is formulated and efficiently solved using the fast iterative shrinkage-thresholding algorithm. The experimental results demonstrated that, compared with the traditional dTV-p algorithm, the proposed method achieved significant improvements in quantitative metrics such as peak signal-to-noise ratio, root mean square error, and structural similarity index, revealing pronounced advantages particularly in suppressing quantum noise and preserving edges.