Metagenomic data is a novel and valuable source for personalized medicine approaches to improve human health. Data Visualization is a crucial technique in data analysis to explore and find patterns in data. Especially, data resources from metagenomic often have very high dimension so humans face big challenges to understand them. In this study, we introduce a visualization method based on Mean-shift algorithm which enables us to observe high-dimensional data via images exhibiting clustered features by the clustering method. Then, these generated synthetic images are fetched into a convolutional neural network to do disease prediction tasks. The proposed method shows promising results when we evaluate the approach on four metagenomic bacterial species abundance datasets related to four diseases including Liver Cirrhosis, Colorectal Cancer, Obesity, and Type 2 Diabetes.