OC-087 THE MESH AND THE MACHINE: DEEP LEARNING TO IDENTIFY HERNIA MESHES ON CT SCANS

Abstract

Abstract Aim To identify the presence of meshes in CT scans of patients who have previously undergone hernia repairs using our novel deep learning algorithms. Materials & Methods We annotated mesh data from 30 ventral hernia CT scans of patients who had undergone previous ventral hernia repairs using computer vision tools. The annotated slices from the CT scans qualified as the ground truth. 80% of the data was used for training and validation. 20% data was used for testing. The novel deep learning model we used had previously identified meshes with lower accuracy levels due to inferior data sets. Results Our deep learning model was able to identify the mesh with an accuracy of 70% when compared to a manual annotator. Retraining our model with more varied mesh data gave us improved results. Conclusions Identification of the mesh is a challenge for surgeons and radiologists. The evolution of deep learning techniques made the identification of obscure structures possible. Meshes present a unique challenge as they constitute a foreign tissue which integrates into native tissue. Our team's novel deep learning algorithms have been effective in identifying meshes, This could play a major role in automated hernia reporting. The surgeon will be benefited from additional information about the plane of mesh and level of integration while planning surgeries for recurrence or mesh explantation.