app.py

import os
import time
import json
import gradio as gr
from gradio_niivueviewer import NiiVueViewer

DEMO_DIR = os.path.dirname(os.path.abspath(__file__))
EXAMPLE_CT   = os.path.join(DEMO_DIR, "input/amos_0004-img.nii.gz")
EXAMPLE_MASK = os.path.join(DEMO_DIR, "output/amos_0004-mask.nii.gz")

# TotalSegmentator labels present in the demo mask
SEG_LABELS = {
    1:  "Spleen",
    2:  "Right Kidney",
    3:  "Left Kidney",
    4:  "Gallbladder",
    5:  "Liver",
    6:  "Stomach",
    7:  "Pancreas",
    8:  "Right Adrenal Gland",
    9:  "Left Adrenal Gland",
    10: "Left Lung (Upper)",
    11: "Left Lung (Lower)",
    13: "Right Lung (Middle)",
    14: "Right Lung (Lower)",
    15: "Esophagus",
    18: "Small Bowel",
    19: "Duodenum",
    20: "Colon",
    21: "Urinary Bladder",
    22: "Prostate/Uterus",
    25: "Sacrum",
    26: "Vertebrae S1",
    27: "Vertebrae L5",
    28: "Vertebrae L4",
    29: "Vertebrae L3",
    30: "Vertebrae L2",
    31: "Vertebrae L1",
    32: "Vertebrae T12",
    33: "Vertebrae T11",
    34: "Vertebrae T10",
    35: "Vertebrae T9",
    36: "Vertebrae T8",
    51: "Left Hip",
    52: "Right Hip",
    63: "Gluteus Maximus (L)",
    64: "Gluteus Maximus (R)",
    65: "Gluteus Medius (L)",
    66: "Gluteus Medius (R)",
    67: "Gluteus Minimus (L)",
    68: "Gluteus Minimus (R)",
    75: "Autochthon (L)",
    76: "Autochthon (R)",
    77: "Iliopsoas (L)",
    78: "Iliopsoas (R)",
    79: "Rib 1 (L)",
    80: "Rib 1 (R)",
    81: "Rib 2 (L)",
    82: "Rib 2 (R)",
    83: "Rib 3 (L)",
    84: "Rib 3 (R)",
    85: "Rib 4 (L)",
    86: "Rib 4 (R)",
    87: "Rib 5 (L)",
    88: "Rib 5 (R)",
    89: "Rib 6 (L)",
    97: "Rib 7 (L)",
    98: "Rib 7 (R)",
    99: "Rib 8 (L)",
    100: "Rib 8 (R)",
    101: "Rib 9 (L)",
    102: "Rib 9 (R)",
    103: "Rib 10 (L)",
    109: "Sternum",
    110: "Costal Cartilage",
    111: "Aorta",
    112: "Pulmonary Vein",
    113: "Superior Vena Cava",
    114: "Inferior Vena Cava",
    115: "Portal/Splenic Vein",
    117: "Heart",
}


def preview_ct(ct_path: str | None):
    """Show CT only (no segmentation yet)."""
    if not ct_path:
        return None
    return [ct_path]


def predict(ct_path: str | None):
    """Fake inference: sleep then return demo mask + label JSON."""
    if not ct_path:
        return None, None
    # Simulate model inference
    time.sleep(4)
    labels_json = {str(k): v for k, v in SEG_LABELS.items()}
    return [ct_path, EXAMPLE_MASK], labels_json


with gr.Blocks(title="Abdominal Organ Segmentation") as demo:
    gr.Markdown(
        "## Abdominal Organ Segmentation Demo\n"
        "Upload a CT scan or click the example below, then press **Run** to segment."
    )

    with gr.Row():
        ct_input = gr.File(
            label="Input CT (.nii / .nii.gz)",
            file_types=[".nii", ".gz"],
            scale=3,
        )
        run_btn = gr.Button("▶  Run Segmentation", variant="primary", scale=1)

    viewer = NiiVueViewer(label="Viewer", seg_labels=SEG_LABELS)
    labels_out = gr.JSON(label="Segmentation Labels")

    gr.Examples(
        examples=[[EXAMPLE_CT]],
        inputs=[ct_input],
        fn=preview_ct,
        outputs=[viewer],
        run_on_click=True,
        label="Example",
    )

    ct_input.change(preview_ct, inputs=ct_input, outputs=viewer)
    run_btn.click(predict, inputs=ct_input, outputs=[viewer, labels_out])


if __name__ == "__main__":
    demo.launch(
        server_port=7870,
        allowed_paths=[DEMO_DIR],
    )