DeepAdipose

Analysis of CT texture features of visceral adipose tissue for evaluation of metabolic disorders and surgery-induced weight loss effects

Citation

Juan Shi¹, Guoqing Bao¹, Jie Hong¹, Simin Wang, Yufei Chen, Shaoqian Zhao, Aibo Gao, Ru Zhang, Jingfen Hu, Wenjie Yang, Fuhua Yan, Ankang Lv, Ruixin Liu, Bin Cui, Yifei Zhang, Weiqiong Gu, Dagan Feng, Weiqing Wang, Jiqiu Wang, Xiuying Wang, Guang Ning, “Deciphering CT texture features of human visceral fat to evaluate metabolic disorders and surgery-induced weight loss effects”, EBiomedicine, vol. 69, p. 103471, 2021, doi: 10.1016/j.ebiom.2021.103471.

¹ Co-first authors.

Full-text paper access from the Lancet: https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(21)00264-4/fulltext

Prerequisites

The following libraries are required:

R, python 3.6, SPSS, SimpleITK, radiomics, matplotlib, pandas, scipy, scikit-learn, tensorflow 1.x and keras

The dataset (Folder "Data")

The dataset contains extracted CT imaging features from 675 volunteer studies and 63 obesity patients (with bariatric surgery).

Features were extracted from umbilical level (allBlockData.xlsx and allSliceData.xlsx)

Major features with routine clinical parameters and metabolic outcomes were provided (all_major_block_features_with_clinical.xlsx)

Features with surgery outcomes were provided in surgery_merged_with_clinical.xlsx

The Code

We provided both python/R source code and corresponding jupyter notebooks.

1. Extract CT features from segmented adipose tissues

After segmentation of visceral adipose tissue (VAT) from CT scans with an in-house software, texture features were extracted from umbilical CT blocks and slices using the following code:

ExtractFeaturesFromBlocksSlices.py (for volunteer cohort)

ExtractFeaturesFromSurgery.py (for surgery cohort)

2. Build deep learning models for evaluation of metabolic outcomes

Deep feed-forward models were built based on major features extracted from CT blocks and slices.

DeepAdiposeBlock.py (with features obtained from CT blocks)

DeepAdiposeSlice.py (with features obtained from CT slices)

3. Evaluate the performance of deep learning models on test data

Load pretrained models from "Results/Models" and evalaute on test data (DeepAdiposeTest.py)

4. Visualization analysis of the correlation between imaging features and metabolic outcomes as well as clinical parameters

Create a Sankey diagram to visualize the correlations (AdiposeSankey.py)

Create heatmaps to visualize the pattern of imaging features in relation to metabolic outcomes (AdiposeHeatmap.r for major features, AdiposeHeatmapFull.r for all features)

ROC analysis of single biomarkers for Metabolic syndrome and insulin resistance (SPSS_ROC_Single_Marker.spv)

5. Surgery analysis with SPSS

Analysis of imaging features in relation to surgery outcomes with SPSS (Surgery_SPSS_Analysis.spv)