This study explores the integration of reinforcement learning (RL) with decision tree models to enhance the classification of acute coronary syndrome (ACS) using 12-lead ECG data. We introduce a novel Q-learning-based RL agent that optimizes feature selection at each node of the decision tree, with the Gini impurity serving as the reward signal. This approach allows the model to dynamically adjust the feature set at each split, potentially improving its decision-making process. To assess the effectiveness of our proposed method, we combined an ensemble of RL-enhanced decision trees with traditional random forest models, creating a hybrid model for ACS classification. The results demonstrate a notable improvement in performance, with the RL-enhanced model achieving an AUROC of 0.83, compared to the 0.80 obtained by standard random forest models. These findings highlight the potential of RL-driven decision trees in improving diagnostic accuracy for ACS, offering a novel and promising approach for further research in AI-augmented medical diagnosis.