The detection of the Diffuse Supernova Neutrino Background (DSNB) flux will provide invaluable insights into constraining cosmological models, core-collapse dynamics and neutrino properties. The Super-Kamiokande-Gd (SK-Gd) experiment currently exhibits the best sensitivity for discovery due to enhanced neutron tagging capability with 0.011% gadolinium sulfate octahydrate, as per this analysis. While the Inverse Beta Decay (IBD) interaction is identifiable in SK-Gd, the low-energy signal is dominated by atmospheric neutrino backgrounds. This study explores a novel approach to background reduction by leveraging topological features of SK events with the discriminative power of Convolutional Neural Networks (CNNs). Well-established techniques for data preprocessing, event selection and feature extraction are used to train CNNs on IBD and atmospheric Neutral Current (NC) Monte-Carlo (MC) events. Preliminary performance of two CNN models highlights the potential of using machine-learning techniques to improve the DSNB signal efficiency