Robust spectrum sensing is crucial for facilitating opportunistic spectrum utilization for secondary users (SU) in the absense of primary users (PU). However, propagation environment factors such as multi-path fading, shadowing, and lack of line of sight (LoS) often adversely affect detection performance. To deal with these issues, this paper focuses on utilizing reconfigurable intelligent surfaces (RIS) to improve spectrum sensing in the scenario wherein both the multi-path fading and noise are correlated. In particular, to leverage the spatially correlated fading, we propose to use maximum eigenvalue detection (MED) for spectrum sensing. We first derive exact distributions of test statistics, i.e., the largest eigenvalue of the sample covariance matrix, observed under the null and signal present hypothesis. Next, utilizing these results, we present the exact closed-form expressions for the false alarm and detection probabilities. In addition, we also optimally configure the phase shift matrix of RIS such that the mean of the test statistics is maximized, thus improving the detection performance. Our numerical analysis demonstrates that the MED's receiving operating characteristic (ROC) curve improves with increased RIS elements, SNR, and the utilization of statistically optimal configured RIS.