Amplitude-frequency-aware deep fusion network for optimal contact selection on STN-DBS electrodes

Parkinson’s disease (PD) is treated effectively by deep brain stimulation (DBS) of the subthalamic nucleus (STN), using an electrode inserted into the head of a PD patient. The electrode has multiple electrical contacts along its length, so the best may be chosen for selectively stimulating the STN. Neurosurgeons usually determine the optimal stimulated contact via the clinical experience of the neurosurgeon and the motor improvement of PD patients. This is a time-consuming and labor-intensive trial-and-error process. The selection of optimal stimulated contact highly depends on the locations of sweet spots, which are manually identified by the characteristic features of microelectrode recordings (MERs). This paper presents an amplitude-frequency-aware deep fusion network for optimal contact selection on STN-DBS electrodes. The method first obtains the amplitude-frequency fusion features by combining the MERs time sequence features and the amplitude sequence features, and then uses the convolutional neural network (CNN) with convolutional block attention module (CBAM) to identify both the border of the STN and the sweet spots to implant the electrode. The optimal stimulated contact can be selected according to the distribution of the sweet spots. Experimental results indicate that, for successful surgeries, neurosurgeons and the proposed AI solution selected the same optimal contacts. Furthermore, the proposed method outperforms the state-of-the-art methods for STN and sweet spot identification. The proposed method shows great potential for optimal contact selection to improve the efficiency of STN-DBS surgery and reduce the dependence on clinicians’ experience.