app.py

import gradio as gr
import spaces
from roi_predictor import predict_rois
from landmark_predictor import predict_landmark, show_landmarks
from cpak_calculator import calculate_cpak

@spaces.GPU
def predict_rois_gpu(image):
    annotated_img, gallery_items, rois = predict_rois(image)
    return annotated_img, gallery_items, rois

@spaces.GPU
def predict_landmark_gpu(image, rois):
    landmarks = predict_landmark(image, rois)
    heatmap_image, heatmap_gallery, landmark_image = show_landmarks(image, rois, landmarks)
    cpak_right_dict, cpak_left_dict = calculate_cpak(image, landmarks)
    
    right_text = (
        f"MPTA: {cpak_right_dict['MPTA']:.2f} °\n"
        f"LDFA: {cpak_right_dict['LDFA']:.2f} °\n"
        f"aHKA: {cpak_right_dict['aHKA']:.2f} ° ({cpak_right_dict['aHKA_label']})\n"
        f"JLO: {cpak_right_dict['JLO']:.2f} ° ({cpak_right_dict['JLO_label']})\n\n"
        f"CPAK: {cpak_right_dict['CPAK']} \n"
    )
    
    left_text = (
        f"MPTA: {cpak_left_dict['MPTA']:.2f} °\n"
        f"LDFA: {cpak_left_dict['LDFA']:.2f} °\n"
        f"aHKA: {cpak_left_dict['aHKA']:.2f} ° ({cpak_left_dict['aHKA_label']})\n"
        f"JLO: {cpak_left_dict['JLO']:.2f} ° ({cpak_left_dict['JLO_label']})\n\n"
        f"CPAK: {cpak_left_dict['CPAK']} \n"
    )
    
    return heatmap_image, heatmap_gallery, landmark_image, right_text, left_text

CUSTOM_CSS = """
.caption-label {color: #888;}
"""

# Define modern Gradio Blocks interface
with gr.Blocks() as demo:
    gr.Markdown(
        """
        # 🦴 Knee Landmarks ROI Detector & Analyzer
        Upload a knee X-ray image to automatically detect regions of interest (ROIs), predict landmarks, and compute CPAK measurements.
        """
    )
    
    with gr.Row():
        # Left Column: Input and Controls
        with gr.Column(scale=1):
            with gr.Tabs():
                with gr.TabItem("Input"):
                    input_image = gr.Image(type="numpy", label="Input Knee X-Ray Image")
            
            
            detect_roi_btn = gr.Button("Detect ROIs", variant="primary")
            predict_landmark_btn = gr.Button("Predict Landmarks", variant="secondary")
            
        # Right Column: Visualizations & Analysis
        with gr.Column(scale=2):
            with gr.Tabs():
                # Tab 1: ROI Detection Output
                with gr.TabItem("1. ROI Detection"):
                    with gr.Row():
                        with gr.Column(scale=1):
                            roi_image = gr.Image(label="Annotated ROIs")
                        with gr.Column(scale=1):
                            roi_gallery = gr.Gallery(label="Cropped Regions", columns=3, height="auto")

                    roi_state = gr.State()

                # Tab 2: Landmarks and Heatmaps
                with gr.TabItem("2. Landmarks"):
                    with gr.Row():
                        with gr.Column(scale=1):
                            heatmap_image = gr.Image(label="Predicted Heatmap")
                        with gr.Column(scale=1):
                            landmark_image = gr.Image(label="Final Landmarks")
                    
                    gr.Markdown("## Landmark Heatmaps")
                    heatmap_gallery = gr.Gallery(label="Heatmap Gallery", columns=4, height="auto")
                
                # Tab 3: Clinical Measurements
                with gr.TabItem("3. CPAK Measurement"):
                    with gr.Row():
                        with gr.Column():
                            gr.Markdown("# Right Side")
                            cpak_right = gr.Textbox(
                                value="",
                                show_label=False,
                                lines=8,
                                interactive=False
                            )
                        with gr.Column():
                            gr.Markdown("# Left Side")
                            cpak_left = gr.Textbox(
                                value="",
                                show_label=False,
                                lines=8,
                                interactive=False
                            )
                    
    detect_roi_btn.click(
        fn=predict_rois_gpu,
        inputs=[input_image],
        outputs=[roi_image, roi_gallery, roi_state]
    )

    predict_landmark_btn.click(
        fn=predict_landmark_gpu,
        inputs=[input_image, roi_state],
        outputs=[heatmap_image, heatmap_gallery, landmark_image, cpak_right, cpak_left]
    )

demo.launch(theme=gr.Theme.from_hub("harsh8001/comic"), css=CUSTOM_CSS, ssr_mode=False)