Efficient handling of mixed-priority, time-critical network requests is essential in systems where reliability and low latency are crucial, such as healthcare and industrial control. In this context, IEEE Time-Sensitive Networking (TSN) standards focus on providing real-time communications on wired Ethernet networks and have been quite mature over the last decade. However, the Internet of medical things (IoMT), requires the same capability of precise timeliness and reliability to be guaranteed over wireless networks. In addition, the underlying networks in a healthcare scenario must support the handling of heterogeneous requests of different priorities for dynamic network conditions. Traditional scheduling approaches often rely on static rules that do not adapt to varying network conditions. This paper presents a Deep Reinforcement Learning (DRL) based adaptive scheduling framework using Proximal Policy Optimization (PPO) to dynamically manage heterogeneous traffic with different priority levels. The proposed agent learns optimal packet scheduling strategies through interaction with a simulated network environment and aims to minimize end-to-end delay, jitter, and packet loss. Experimental results show that the PPO-based scheduler achieves up to 30% lower latency and 55% lower jitter compared to static and DQN-based schedulers, demonstrating its effectiveness for real-time, priority-aware communication.