Ovarian Classification

Code for paper Deep Learning Models to Classify Benign and Malignant Ovarian Lesions on MRI: A Multicenter Study

Our study aims to develop a comprehensive end-to-end pipeline. We first evaluate SAM's performance in segmenting ovarian lesions on MRI. Building on this, we then develop a robust DL model for distinguishing between benign and malignant ovarian lesions using MRI data. By leveraging SAM-segmented images, we seek to enhance the effectiveness and generalizability of our diagnostic approach, ultimately improving clinical outcomes for patients. The workflow depicting the integration of T1CE and T2WI with ROI segmentation, pre-processed along with clinical data, into the deep learning model for lesion classification.

Note:

./code and ./dataset contain the code for each component of the Ovarian classification model and pre-process code for the both T1 and T2 dataset. These programs cannot be run with the full T1/T2 ovarian dataset, but help illustrate the exact steps we did to form our projects.

./3. T1-2 classification model-with-clinical-dymlp-dual contains the model weights for the ovarian classification model. See the respective README.md in the subfolder to see how to run each component.

Dependencies:

Using Python version of python=3.10. To establish the environment, run this code in the shell:

conda activate ovarian_classcify
pip install -e requrements.txt.

That will create the environment ovarian classification model we used.

Usage

Environment setup

Activate the environment by running

conda activate ovarian_classcify

Code pipeline: data preprocessing, SAM model, and classification model inference

Data preprocess:

# Run the preprocessed code
python classification-data-preprocessing.py

Data preprocess files and code organizations:

- T1-ROI-preprocess & T2-ROI-preprocess: Folder that stores the preprocessed T1/T2 files. This folder will fill after running classification-data-preprocessing.py
- label-classifyT1C+T2-clear.xlsx: Template for what this xlsx data file should contain with all column headers.
- label-T1C+T2-clear.xlsx: Template for what this xlsx data file should contain with all column headers. 

SAM model:

We use the Segment Anything Model (SAM) to aid the segmentation of 3D data from tomography or other imaging techniques. Detailed instructions on how to use this model can be found here.

classification model inference

# Run the model inference code
python ovarian classification-dualbranch-fusion-test.py

• This code takes in the input of the classification data along with the preprocessed T1 and T2 image data to output classification. The end of the model will print evaluation metrics of accuracy.
• When running this model, you Will need to change all the “#Directory Address” portions but have currently inputted templates of data path locations
• When running this model, you Will need to change all of the array endpoints throughout the code for test images, labels, and clinic data based on how data is formatted

Citation

If you find our work helpful, please cite us