Predictive and Explainable Artificial Intelligence for Weight Loss After Sleeve Gastrectomy: Insights from Wide and Deep Learning with Medical Image and Non-Image Data
There has been no feasible approach for predicting weight loss after bariatric surgery. This study develops wide and deep learning (WAD), a predictive and explainable artificial intelligence for weight loss after sleeve gastrectomy with medical image and non-image data, such as electronic medical re...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-02-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/15/5/2457 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There has been no feasible approach for predicting weight loss after bariatric surgery. This study develops wide and deep learning (WAD), a predictive and explainable artificial intelligence for weight loss after sleeve gastrectomy with medical image and non-image data, such as electronic medical records (EMRs). Prospective data came from 42 patients with sleeve gastrectomy at a university hospital. They were followed for one year after surgery. The dependent variable consisted of three categories: minimal, moderate, and significant change groups, classified based on postoperative percentage total weight loss (%TWL) in body mass index. A pair of 100 images and their non-image data came from each patient, with 4200 pairs from 42 patients in total. A WAD model was trained and tested with 3200 and 1000 pairs, respectively. Here, the WAD model combined a convolutional neural network (CNN) for image data and a linear layer for non-image data (EMR). The study population included 42 patients, with a mean age of 36.6 years (standard deviation SD 11.0) and a female proportion of 58% (26/45). On average, %TWL was 19.1 (SD 2.8), 27.3 (SD 2.2), and 35.1 (SD 4.7) for the minimal, moderate, and significant change groups, respectively. The corresponding accuracy outcomes were 61%, 100%, and 75% for the minimal, moderate, and significant change groups (average 71%). When the minimal and moderate change groups were combined, the accuracy was 100% for the combined group and 75% for the significant change group, with an overall average accuracy of 88%. Baseline HOMA2-B, insulin, and vitamin B12 were major predictors of %TWL. The optimal region of interest for predicting %TWL was found to be the entire cross-section above the diaphragm. In conclusion, WAD is an effective predictive and explainable artificial intelligence for weight loss following sleeve gastrectomy with image and non-image data. The most important predictors of postoperative weight loss were identified as baseline HOMA2-B, insulin, and vitamin B12 levels, while the key region of interest (ROI) in abdominal CT imaging was the entire cross-section located above the diaphragm. |
|---|---|
| ISSN: | 2076-3417 |