STT-BOX v1.0

Example of visualization codes for the article "A deep learning-based system trained for gastrointestinal stromal tumor screening can identify multiple types of soft tissue tumors" published in The American Journal of Pathology.

Environments and Requirements

• Ubuntu 16.04.6 LTS
• CPU Intel(R) Xeon(R) Gold 5218 CPU @ 2.30GHz
• GPU NVIDIA Tesla T4 16G
• CUDA 10.0.130
• Pytorch 1.2.0
• Python 3.7.6
• Numpy 1.18.1
• Openslide
• Mathplotlib
• PIL
• opencv (cv2)
• skimage

Implementation

1) Prepare the image waiting for inference

Please prepare the Whole-Slide Image (WSI) for inference. The file format of the image should be ".svs", ".tiff", ".ndpi", and etc. due to th usage of OpenSlide. Then, modify the parameter for image path in inference_STT-BOX.py.

# inference_STT-BOX.py
class Option(object):
    def __init__(self):
        self.img_path = './your_image_name.svs' # example of the image path 

2) Pre-procss the trained model

After decompressing the five compressed packages in the folder "model", you will receive a file named "STT_BOX_v1.pth". Please place this file in "./model/STT_BOX_v1.pth".

3) Obtain the output heatmap

Run the file inference_STT-BOX.py. If a specific GPU is selected (such as No. 3), run CUDA_VISIBLE_DEVICES=3 python inference_STT-BOX.py. Finally, you will obtain the heatmap, e.g., "your_file_name_heatmap.png". Note that the codes are just examples to obtain visualization, and different outputs will be gained when the parameters are changed.

Citation

Please cite:

Meng Z, Wang G, Su F, et al. A deep learning-based system trained for gastrointestinal stromal tumor screening can identify multiple types of soft tissue tumors[J]. The American Journal of Pathology, 2023.

@article{meng2023deep,
title={A deep learning-based system trained for gastrointestinal stromal tumor screening can identify multiple types of soft tissue tumors}, author={Meng, Zhu and Wang, Guangxi and Su, Fei and Liu, Yan and Wang, Yuxiang and Yang, Jing and Luo, Jianyuan and Cao, Fang and Zhen, Panpan and Huang, Binhua and others},
journal={The American Journal of Pathology},
year={2023},
publisher={Elsevier}
}