This study proposes a multiscale enhanced graph convolutional network (MSE-GCN) framework integrating phenotypic and functional MRI data for autism spectrum disorder (ASD) detection. Population graphs represent individuals as vertices, weighted using fuzzy inference on phenotypic data. Random walks and parallel GCN embeddings capture complex patterns, followed by multilayer perceptron for feature selection. Applied to the ABIDE dataset, the method achieved 87% accuracy, outperforming existing approaches, and demonstrates the benefit of combining multimodal imaging with graph-based deep learning for ASD detection.