Patients with drug-resistant epilepsy can be treated with the Responsive Neurostimulation (RNS) system. This system has the ability to sense the electrical activity of the brain and provide stimulation when epileptiform patterns are detected. Patients implanted with the RNS system in the thalamus experience longer periods of stimulation. Since the implanted electrodes have a dual function of sensing and stimulating, patients with RNS in the thalamus show recordings with significant data information loss. Consequently, deep learning algorithms developed to analyze the presence of epileptogenic patterns in RNS recordings have shown decreased performance when evaluated in thalamic patients. This study aims to enhance the detection capability of epileptogenic patterns in a deep neural network (iESPnet) for thalamic patients by integrating a spatial attention mechanism called Dynamic Spatial Filtering (DSF) into the learning process. Various experiments were conducted to determine if the combined use of DSF and iESPnet has the potential to improve the detection capacity of epileptogenic patterns. A database of 30 patients implanted with RNS was utilized. The results demonstrate that the integration of the attention mechanism has the potential to enhance the network’s detection capabilities, achieving relative improvements of up to 30% compared to the original network architecture. Thus, this work defines an explorative path to improve personalized treatment for patients with refractory epilepsy through the automation of epileptogenic pattern detection, significantly reducing the workload of specialists.