<p>Nowadays, cancer is one of the fatal diseases with an increasing death toll globally. Plenteous types of treatments have emerged to remove tumors. So far, auto-detection of trivial-size tumors is quite challenging for image-guided radiation therapy (IGRT) systems. Computer vision-based techniques are being applied with imaged-guided radiation therapy in the case of tiny tumor treatment, comparatively with more precision and accuracy. These techniques are objectively designed to assist the IGRT team in diagnosing it accurately and improving outcome predictions. This research paper proposes a deep reinforcement learning-based scheme to detect tiny size tumors autonomously. Using this technique, an image-guided radiation therapy system, like cyber-knife, may learn to detect tumors without any human in- intervention. A composite technique of deep Q-Network and Genetic Algorithm (GA), Open AI, and ROS frameworks are used as building blocks; with 2-D simulation, prominent results are produced. Around 95% of detection accuracy is obtained by using this proposed solution, which gradually converges optimality by learning reward. The previous state of art achieves more than 93% accuracy. The promised outcome endorses the idea that deep reinforcement learning with evolutionary algorithms may enable IGRT systems to detect tumors autonomously with more precision and accuracy.</p>