app.py

import spaces

import tempfile
from pathlib import Path

import SimpleITK as sitk
import torch
from mrsegmentator import inference
from mrsegmentator.utils import add_postfix

import gradio as gr
import utils


description_markdown = """
- **GitHub: https://github.com/hhaentze/mrsegmentator
- **Paper: https://arxiv.org/abs/2405.06463"
- **Please Note:** This tool is intended for research purposes only.
"""


css = """

h1 {
    text-align: center;
    display:block;
}
.markdown-block {
    background-color: #0b0f1a; /* Light gray background */
    color: white;             /* Black text */
    padding: 10px;            /* Padding around the text */
    border-radius: 5px;       /* Rounded corners */
    box-shadow: 0 0 10px rgba(11,15,26,1);
    display: inline-flex;       /* Use inline-flex to shrink to content size */
    flex-direction: column;
    justify-content: center;    /* Vertically center content */
    align-items: center;        /* Horizontally center items within */
    margin: auto;               /* Center the block */
}

.markdown-block ul, .markdown-block ol {
    background-color: #1e2936;
    border-radius: 5px;
    padding: 10px;
    box-shadow: 0 0 10px rgba(0,0,0,0.3);
    padding-left: 20px;         /* Adjust padding for bullet alignment */
    text-align: left;           /* Ensure text within list is left-aligned */
    list-style-position: inside;/* Ensures bullets/numbers are inside the content flow */
}

footer {
    display:none !important
}
"""

examples = ["amos_0555.nii.gz","amos_0517.nii.gz", "amos_0541.nii.gz", "amos_0571.nii.gz"]


def save_file(segmentation, path):
    """If the segmentation comes from our sample files directly return the path.
    Otherwise save it to the temporary file that was previously allocated by the input image"""

    if Path(path).name in examples:
        path = "segmentations/" + add_postfix(path, "seg")
    else:
        sitk.WriteImage(segmentation, path)

    return path


@spaces.GPU(duration=150)
def infer(image_path):
    with tempfile.TemporaryDirectory() as tmpdirname:

        if torch.cuda.is_available():
            inference.infer([image_path], tmpdirname, [0, 1, 2, 3, 4], cpu_only=False, split_level=1)
        else:
            inference.infer([image_path], tmpdirname, [0], cpu_only=True, split_level=1)
        filename = add_postfix(Path(image_path).name, "seg")
        segmentation = sitk.ReadImage(tmpdirname + "/" + filename)

    return segmentation


def infer_wrapper(input_file, image_state, seg_state, slider=50):

    filename = Path(input_file).name

    # inference
    if filename in examples:
        segmentation = sitk.ReadImage("segmentations/" + add_postfix(filename, "seg"))
    else:
        segmentation = infer(input_file.name)

    # save file
    seg_path = save_file(segmentation, input_file.name)
    seg_state.append(utils.sitk2numpy(segmentation))

    return utils.display(image_state[-1], seg_state[-1], slider), seg_state, seg_path


with gr.Blocks(css=css, title="MRSegmentator") as iface:

    gr.Markdown("# Robust Multi-Modality Segmentation of 40 Classes in MRI and CT Imaging")
    gr.Markdown(description_markdown, elem_classes="markdown-block")

    image_state = gr.State([])
    seg_state = gr.State([])

    with gr.Row():
        with gr.Column():

            input_file = gr.File(
                type="filepath", label="Upload an MRI Image (.nii/.nii.gz)", file_types=[".gz", ".nii.gz"]
            )
            gr.Examples(["images/" + ex for ex in examples], input_file)

            with gr.Row():
                submit_button = gr.Button("Run", variant="primary")
                clear_button = gr.ClearButton()

            slider = gr.Slider(1, 100, value=50, step=2, label="Select (relative) Slice")
            download_file = gr.File(label="Download Segmentation", interactive=False)

        with gr.Column():
            overlay_image_np = gr.AnnotatedImage(label="Axial View")

    input_file.change(
        utils.read_and_display,
        inputs=[input_file, image_state, seg_state],
        outputs=[overlay_image_np, image_state, seg_state],
    )
    slider.change(utils.display, inputs=[image_state, seg_state, slider], outputs=[overlay_image_np])

    submit_button.click(
        infer_wrapper,
        inputs=[input_file, image_state, seg_state, slider],
        outputs=[overlay_image_np, seg_state, download_file],
    )

    clear_button.add([input_file, overlay_image_np, image_state, seg_state, download_file])


if __name__ == "__main__":
    iface.queue()
    iface.launch()