OC-061 IMAGE-BASED DEEP LEARNING MODELS (DLMS) IDENTIFY ABDOMINAL WALL RECONSTRUCTION (AWR) PATIENTS RECEIVING PREOPERATIVE BOTULINUM TOXIN A (BTA) WHO REQUIRE COMPONENT SEPARATION TECHNIQUE (CST)

Abstract

Abstract Aim Preoperative BTA facilitates muscle/fascia elongation and fascial closure, which increases the durability of hernia repair and may reduce requirement for CST. The aim of this study was to develop image-based DLMs that predict whether patient will require CST if injected with BTA prior to surgery. Methods An institutional database was used to identify AWR patients who received preoperative BTA and had preoperative CT imaging. Axial CT cuts of the hernias were rendered to train and develop a DLM. The primary outcome was a ROC for predicting CST. The DLM was tested on CT scans from a pre-identified subset of patients who underwent CST and did not receive preoperative BTA. Results There were 116 patients who met criteria (4,580 CT images). Of these patients, 69 (59.5%) required CST (2884 images); 47 patients (40.5%) did not undergo CST (1,696 images). The DLM ROC was 0.78 (Figure 1); accuracy, sensitivity, and specificity were 0.79, 0.86, and 0.68, respectively. There were 98 patients in the test set; 57 (58.1%) were predicted to require CST and 41 (41.9%) were not. For patients with an M1 hernia component, 21.1% were predicted to be spared CST versus 46.8% those with an M2-M5 hernia component (p=0.04). Figure 1 Model Performance for CST Prediction Conclusions Image-based DLMs accurately predicted which patients receiving preoperative BTA may require CST. When the DLM was applied to a test set, patients with an M1 hernias were very likely to require CST, which is consistent with previously reported data, further validating the model.