Spatial transcriptomics has emerged as a transformative technology for mapping gene expression within tissue contexts, offering unprecedented insights into disease mechanisms. However, extracting actionable insights from these high-dimensional datasets remains challenging due to their complexity and noise. In this paper, we propose a novel framework that integrates spatial transcriptomics with advanced computer vision techniques to identify therapeutic targets in drug discovery. Our approach leverages deep learning-based segmentation and graph neural networks (GNNs) to capture spatial relationships and enhance interpretability. Experiments on benchmark datasets demonstrate significant improvements in identifying disease-specific biomarkers compared to traditional methods. This work underscores the potential of computer vision to revolutionize drug discovery by enabling faster and more accurate target identification.