Deep Learning-assisted Surgical Margin Assessment for Canine Soft Tissue Sarcoma Based on Polarization-sensitive Optical Coherence Tomography

This repository contains the official PyTorch implementation of the following paper:

Deep Learning-assisted Surgical Margin Assessment for Canine Soft Tissue Sarcoma Based on Polarization-sensitive Optical Coherence Tomography

Yuanlong Wang, Laura E. Selmic, Ping Zhang

Abstract: Soft tissue sarcomas (STS) are malignant locally aggressive neoplasms arising from mesenchymal cells. The typical treatment for canine STS is surgical removals. While post-operative margin assessment is traditionally used, optical coherence tomography (OCT) has been reported for intra-operative surgical margin assessment. Polarization-sensitive OCT (PS-OCT) has shown the potential to differentiate tissue types in human breast cancer surgical margin imaging. In this paper, we report a method based on deep learning for fast and accurate intra-operative OCT assessment for paired SD-OCT and PS-OCT images. We tested our proposed method on a canine STS surgical margin dataset and the result showed 0.989 Area Under the Receiver Operating Curve (AUROC) and 91% accuracy in detecting positive margins. We also conducted a case study that used our model to generate diagnosis curves to locate cancerous areas in OCT images. The case study further validates the effectiveness and reliability of our method.

Framework

We implement models with two fusion strategies to conduct margin assessment for PS-OCT images, namely duplicate backbone and channel merge. Our models are based on ResNet for image feature extractors. Their structures are shown in the figure below.

Files directory

./
│
├─assets    # assets for README.md
├─data      # place to put image files
├─dataset
│  │  __init__.py
│  └─ imgset.py
├─model
│  │  __init__.py
│  └─ ResNet.py          
│  main.py                  # script for classification training and testing
│  PS_image_preprocess.py   # crop images into patches 
│  README.md
└─ test.py                  # generate diagnosis curve 

Requirements

Python 3.7
PyTorch 1.13.1
pandas 1.3.5
scikit-learn 1.0.2

Code Usage

Data is available upon request from the corresponding author

1. put the original data into ./data/raw_ps
2. run PS_image_preprocess.py
3. run main.py, with the following arguments:

optional arguments:
  -h, --help            show this help message and exit
  -e EPOCH, --epoch EPOCH
                        number of epoch to train
  -m MODE, --mode MODE  image fusion mode, select from dup_backbone and channel_merge
  -a, --augmentation    use data augmentation method
  --keys KEYS           psoct measurements to use, should be certain combination of "d" for DOPU, "o" for Optial Axis, and "r" for Retardation

4. run test.py to generate prediction curve for trained models, with the following arguments:

  -h, --help            show this help message and exit
  -m MODE, --mode MODE  image fusion mode, select from dup_backbone and channel_merge
  -a, --augmentation    use the model trained with data augmentation method
  --keys KEYS           use the model trained with psoct measurements in KEYS