import gradio as gr from app import demo as app import os _docs = {'MedicalImageAnalyzer': {'description': 'A Gradio component for AI-agent compatible medical image analysis.\n\nProvides structured output for:\n- HU value analysis (CT only)\n- Tissue classification\n- Fat segmentation (subcutaneous, visceral)\n- Confidence scores and reasoning', 'members': {'__init__': {'value': {'type': 'typing.Optional[typing.Dict[str, typing.Any]][\n typing.Dict[str, typing.Any][str, typing.Any], None\n]', 'default': 'None', 'description': None}, 'label': {'type': 'typing.Optional[str][str, None]', 'default': 'None', 'description': None}, 'info': {'type': 'typing.Optional[str][str, None]', 'default': 'None', 'description': None}, 'every': {'type': 'typing.Optional[float][float, None]', 'default': 'None', 'description': None}, 'show_label': {'type': 'typing.Optional[bool][bool, None]', 'default': 'None', 'description': None}, 'container': {'type': 'typing.Optional[bool][bool, None]', 'default': 'None', 'description': None}, 'scale': {'type': 'typing.Optional[int][int, None]', 'default': 'None', 'description': None}, 'min_width': {'type': 'typing.Optional[int][int, None]', 'default': 'None', 'description': None}, 'visible': {'type': 'typing.Optional[bool][bool, None]', 'default': 'None', 'description': None}, 'elem_id': {'type': 'typing.Optional[str][str, None]', 'default': 'None', 'description': None}, 'elem_classes': {'type': 'typing.Union[typing.List[str], str, NoneType][\n typing.List[str][str], str, None\n]', 'default': 'None', 'description': None}, 'render': {'type': 'typing.Optional[bool][bool, None]', 'default': 'None', 'description': None}, 'key': {'type': 'typing.Union[int, str, NoneType][int, str, None]', 'default': 'None', 'description': None}, 'analysis_mode': {'type': 'str', 'default': '"structured"', 'description': '"structured" for AI agents, "visual" for human interpretation'}, 'include_confidence': {'type': 'bool', 'default': 'True', 'description': 'Include confidence scores in results'}, 'include_reasoning': {'type': 'bool', 'default': 'True', 'description': 'Include reasoning/explanation for findings'}, 'segmentation_types': {'type': 'typing.List[str][str]', 'default': 'None', 'description': 'List of segmentation types to perform'}}, 'postprocess': {'value': {'type': 'typing.Dict[str, typing.Any][str, typing.Any]', 'description': None}}, 'preprocess': {'return': {'type': 'typing.Dict[str, typing.Any][str, typing.Any]', 'description': None}, 'value': None}}, 'events': {'change': {'type': None, 'default': None, 'description': 'Triggered when the value of the MedicalImageAnalyzer changes either because of user input (e.g. a user types in a textbox) OR because of a function update (e.g. an image receives a value from the output of an event trigger). See `.input()` for a listener that is only triggered by user input.'}, 'select': {'type': None, 'default': None, 'description': 'Event listener for when the user selects or deselects the MedicalImageAnalyzer. Uses event data gradio.SelectData to carry `value` referring to the label of the MedicalImageAnalyzer, and `selected` to refer to state of the MedicalImageAnalyzer. See EventData documentation on how to use this event data'}, 'upload': {'type': None, 'default': None, 'description': 'This listener is triggered when the user uploads a file into the MedicalImageAnalyzer.'}, 'clear': {'type': None, 'default': None, 'description': 'This listener is triggered when the user clears the MedicalImageAnalyzer using the clear button for the component.'}}}, '__meta__': {'additional_interfaces': {}, 'user_fn_refs': {'MedicalImageAnalyzer': []}}} abs_path = os.path.join(os.path.dirname(__file__), "css.css") with gr.Blocks( css=abs_path, theme=gr.themes.Default( font_mono=[ gr.themes.GoogleFont("Inconsolata"), "monospace", ], ), ) as demo: gr.Markdown( """ # `gradio_medical_image_analyzer`
PyPI - Version Static Badge Static Badge
AI-agent optimized medical image analysis component for Gradio with DICOM support """, elem_classes=["md-custom"], header_links=True) app.render() gr.Markdown( """ ## Installation ```bash pip install gradio_medical_image_analyzer ``` ## Usage ```python #!/usr/bin/env python3 \"\"\" Demo for MedicalImageAnalyzer - Enhanced with file upload and overlay visualization \"\"\" import gradio as gr import numpy as np import sys import os import cv2 from pathlib import Path # Add backend to path sys.path.insert(0, os.path.join(os.path.dirname(os.path.dirname(__file__)), 'backend')) from gradio_medical_image_analyzer import MedicalImageAnalyzer def draw_roi_on_image(image, roi_x, roi_y, roi_radius): \"\"\"Draw ROI circle on the image\"\"\" # Convert to RGB if grayscale if len(image.shape) == 2: image_rgb = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB) else: image_rgb = image.copy() # Draw ROI circle center = (int(roi_x), int(roi_y)) radius = int(roi_radius) # Draw outer circle (white) cv2.circle(image_rgb, center, radius, (255, 255, 255), 2) # Draw inner circle (red) cv2.circle(image_rgb, center, radius-1, (255, 0, 0), 2) # Draw center cross cv2.line(image_rgb, (center[0]-5, center[1]), (center[0]+5, center[1]), (255, 0, 0), 2) cv2.line(image_rgb, (center[0], center[1]-5), (center[0], center[1]+5), (255, 0, 0), 2) return image_rgb def create_fat_overlay(base_image, segmentation_results): \"\"\"Create overlay image with fat segmentation highlighted\"\"\" # Convert to RGB if len(base_image.shape) == 2: overlay_img = cv2.cvtColor(base_image, cv2.COLOR_GRAY2RGB) else: overlay_img = base_image.copy() # Check if we have segmentation masks if not segmentation_results or 'segments' not in segmentation_results: return overlay_img segments = segmentation_results.get('segments', {}) # Apply subcutaneous fat overlay (yellow) if 'subcutaneous' in segments and segments['subcutaneous'].get('mask') is not None: mask = segments['subcutaneous']['mask'] yellow_overlay = np.zeros_like(overlay_img) yellow_overlay[mask > 0] = [255, 255, 0] # Yellow overlay_img = cv2.addWeighted(overlay_img, 0.7, yellow_overlay, 0.3, 0) # Apply visceral fat overlay (red) if 'visceral' in segments and segments['visceral'].get('mask') is not None: mask = segments['visceral']['mask'] red_overlay = np.zeros_like(overlay_img) red_overlay[mask > 0] = [255, 0, 0] # Red overlay_img = cv2.addWeighted(overlay_img, 0.7, red_overlay, 0.3, 0) # Add legend cv2.putText(overlay_img, "Yellow: Subcutaneous Fat", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 0), 2) cv2.putText(overlay_img, "Red: Visceral Fat", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 0, 0), 2) return overlay_img def process_and_analyze(file_obj, modality, task, roi_x, roi_y, roi_radius, symptoms, show_overlay=False): \"\"\" Processes uploaded file and performs analysis \"\"\" if file_obj is None: return None, "No file selected", None, {}, None # Create analyzer instance analyzer = MedicalImageAnalyzer( analysis_mode="structured", include_confidence=True, include_reasoning=True ) try: # Process the file (DICOM or image) file_path = file_obj.name if hasattr(file_obj, 'name') else str(file_obj) pixel_array, display_array, metadata = analyzer.process_file(file_path) # Update modality from file metadata if it's a DICOM if metadata.get('file_type') == 'DICOM' and 'modality' in metadata: modality = metadata['modality'] # Prepare analysis parameters analysis_params = { "image": pixel_array, "modality": modality, "task": task } # Add ROI if applicable if task in ["analyze_point", "full_analysis"]: # Scale ROI coordinates to image size h, w = pixel_array.shape roi_x_scaled = int(roi_x * w / 512) # Assuming slider max is 512 roi_y_scaled = int(roi_y * h / 512) analysis_params["roi"] = { "x": roi_x_scaled, "y": roi_y_scaled, "radius": roi_radius } # Add clinical context if symptoms: analysis_params["clinical_context"] = {"symptoms": symptoms} # Perform analysis results = analyzer.analyze_image(**analysis_params) # Create visual report visual_report = create_visual_report(results, metadata) # Add metadata info info = f"📄 {metadata.get('file_type', 'Unknown')} | " info += f"🏥 {modality} | " info += f"📐 {metadata.get('shape', 'Unknown')}" if metadata.get('window_center'): info += f" | Window C:{metadata['window_center']:.0f} W:{metadata['window_width']:.0f}" # Create overlay image if requested overlay_image = None if show_overlay: # For ROI visualization if task in ["analyze_point", "full_analysis"] and roi_x and roi_y: overlay_image = draw_roi_on_image(display_array.copy(), roi_x_scaled, roi_y_scaled, roi_radius) # For fat segmentation overlay (simplified version since we don't have masks in current implementation) elif task == "segment_fat" and 'segmentation' in results and modality == 'CT': # For now, just draw ROI since we don't have actual masks overlay_image = display_array.copy() if len(overlay_image.shape) == 2: overlay_image = cv2.cvtColor(overlay_image, cv2.COLOR_GRAY2RGB) # Add text overlay about fat percentages if 'statistics' in results['segmentation']: stats = results['segmentation']['statistics'] cv2.putText(overlay_image, f"Total Fat: {stats.get('total_fat_percentage', 0):.1f}%", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) cv2.putText(overlay_image, f"Subcutaneous: {stats.get('subcutaneous_fat_percentage', 0):.1f}%", (10, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 0), 2) cv2.putText(overlay_image, f"Visceral: {stats.get('visceral_fat_percentage', 0):.1f}%", (10, 90), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 0, 0), 2) return display_array, info, visual_report, results, overlay_image except Exception as e: error_msg = f"Error: {str(e)}" return None, error_msg, f"
❌ {error_msg}
", {"error": error_msg}, None def create_visual_report(results, metadata): \"\"\"Creates a visual HTML report with improved styling\"\"\" html = f\"\"\"

🏥 Medical Image Analysis Report

📋 Metadata

File Type: {metadata.get('file_type', 'Unknown')}
Modality: {results.get('modality', 'Unknown')}
Image Size: {metadata.get('shape', 'Unknown')}
Timestamp: {results.get('timestamp', 'N/A')}
\"\"\" # Point Analysis if 'point_analysis' in results: pa = results['point_analysis'] tissue = pa.get('tissue_type', {}) html += f\"\"\"

🎯 Point Analysis

\"\"\" if results.get('modality') == 'CT': html += f\"\"\" \"\"\" else: html += f\"\"\" \"\"\" html += f\"\"\"
Position: ({pa.get('location', {}).get('x', 'N/A')}, {pa.get('location', {}).get('y', 'N/A')})
HU Value: {pa.get('hu_value', 'N/A'):.1f}
Intensity: {pa.get('intensity', 'N/A'):.3f}
Tissue Type: {tissue.get('icon', '')} {tissue.get('type', 'Unknown').replace('_', ' ')}
Confidence: {pa.get('confidence', 'N/A')}
\"\"\" if 'reasoning' in pa: html += f\"\"\"

💭 {pa['reasoning']}

\"\"\" html += "
" # Segmentation Results if 'segmentation' in results and results['segmentation']: seg = results['segmentation'] if 'statistics' in seg: # Fat segmentation for CT stats = seg['statistics'] html += f\"\"\"

🔬 Fat Segmentation Analysis

Total Fat

{stats.get('total_fat_percentage', 0):.1f}%

Subcutaneous

{stats.get('subcutaneous_fat_percentage', 0):.1f}%

Visceral

{stats.get('visceral_fat_percentage', 0):.1f}%

V/S Ratio

{stats.get('visceral_subcutaneous_ratio', 0):.2f}

\"\"\" if 'interpretation' in seg: interp = seg['interpretation'] obesity_color = "#16a34a" if interp.get("obesity_risk") == "normal" else "#d97706" if interp.get("obesity_risk") == "moderate" else "#dc2626" visceral_color = "#16a34a" if interp.get("visceral_risk") == "normal" else "#d97706" if interp.get("visceral_risk") == "moderate" else "#dc2626" html += f\"\"\"

Risk Assessment

Obesity Risk: {interp.get('obesity_risk', 'N/A').upper()}
Visceral Risk: {interp.get('visceral_risk', 'N/A').upper()}
\"\"\" if interp.get('recommendations'): html += \"\"\"
💡 Recommendations
    \"\"\" for rec in interp['recommendations']: html += f"
  • {rec}
    • " html += "
" html += "
" html += "
" # Quality Assessment if 'quality_metrics' in results: quality = results['quality_metrics'] quality_colors = { 'excellent': '#16a34a', 'good': '#16a34a', 'fair': '#d97706', 'poor': '#dc2626', 'unknown': '#6b7280' } q_color = quality_colors.get(quality.get('overall_quality', 'unknown'), '#6b7280') html += f\"\"\"

📊 Image Quality Assessment

Overall Quality: {quality.get('overall_quality', 'unknown').upper()}
\"\"\" if quality.get('issues'): html += f\"\"\"
Issues Detected:
    \"\"\" for issue in quality['issues']: html += f"
  • {issue}
    • " html += "
" html += "
" html += "
" return html def create_demo(): with gr.Blocks( title="Medical Image Analyzer - Enhanced Demo", theme=gr.themes.Soft( primary_hue="blue", secondary_hue="blue", neutral_hue="slate", text_size="md", spacing_size="md", radius_size="md", ).set( # Medical blue theme colors body_background_fill="*neutral_950", body_background_fill_dark="*neutral_950", block_background_fill="*neutral_900", block_background_fill_dark="*neutral_900", border_color_primary="*primary_600", border_color_primary_dark="*primary_600", # Text colors for better contrast body_text_color="*neutral_100", body_text_color_dark="*neutral_100", body_text_color_subdued="*neutral_300", body_text_color_subdued_dark="*neutral_300", # Button colors button_primary_background_fill="*primary_600", button_primary_background_fill_dark="*primary_600", button_primary_text_color="white", button_primary_text_color_dark="white", ), css=\"\"\" /* Medical blue theme with high contrast */ :root { --medical-blue: #1e40af; --medical-blue-light: #3b82f6; --medical-blue-dark: #1e3a8a; --text-primary: #f9fafb; --text-secondary: #e5e7eb; --bg-primary: #0f172a; --bg-secondary: #1e293b; --bg-tertiary: #334155; } /* Override default text colors for medical theme */ * { color: var(--text-primary) !important; } /* Style the file upload area */ .file-upload { border: 2px dashed var(--medical-blue-light) !important; border-radius: 8px !important; padding: 20px !important; text-align: center !important; background: var(--bg-secondary) !important; transition: all 0.3s ease !important; color: var(--text-primary) !important; } .file-upload:hover { border-color: var(--medical-blue) !important; background: var(--bg-tertiary) !important; box-shadow: 0 0 20px rgba(59, 130, 246, 0.2) !important; } /* Ensure report text is readable with white background */ .medical-report { background: #ffffff !important; border: 2px solid var(--medical-blue-light) !important; border-radius: 8px !important; padding: 16px !important; color: #1a1a1a !important; } .medical-report * { color: #1f2937 !important; /* Dark gray text */ } .medical-report h2 { color: #1e40af !important; /* Medical blue for main heading */ } .medical-report h3, .medical-report h4 { color: #1e3a8a !important; /* Darker medical blue for subheadings */ } .medical-report strong { color: #374151 !important; /* Darker gray for labels */ } .medical-report td { color: #1f2937 !important; /* Ensure table text is dark */ } /* Report sections with light blue background */ .medical-report > div { background: #f0f9ff !important; color: #1f2937 !important; } /* Medical blue accents for UI elements */ .gr-button-primary { background: var(--medical-blue) !important; border-color: var(--medical-blue) !important; } .gr-button-primary:hover { background: var(--medical-blue-dark) !important; border-color: var(--medical-blue-dark) !important; } /* Tab styling */ .gr-tab-item { border-color: var(--medical-blue-light) !important; } .gr-tab-item.selected { background: var(--medical-blue) !important; color: white !important; } /* Accordion styling */ .gr-accordion { border-color: var(--medical-blue-light) !important; } /* Slider track in medical blue */ input[type="range"]::-webkit-slider-track { background: var(--bg-tertiary) !important; } input[type="range"]::-webkit-slider-thumb { background: var(--medical-blue) !important; } \"\"\" ) as demo: gr.Markdown(\"\"\" # 🏥 Medical Image Analyzer Supports **DICOM** (.dcm) and all image formats with automatic modality detection! \"\"\") with gr.Row(): with gr.Column(scale=1): # File upload - no file type restrictions with gr.Group(): gr.Markdown("### 📤 Upload Medical Image") file_input = gr.File( label="Select Medical Image File (.dcm, .dicom, IM_*, .png, .jpg, etc.)", file_count="single", type="filepath", elem_classes="file-upload" # Note: NO file_types parameter = accepts ALL files ) gr.Markdown(\"\"\" Accepts: DICOM (.dcm, .dicom), Images (.png, .jpg, .jpeg, .tiff, .bmp), and files without extensions (e.g., IM_0001, IM_0002, etc.) \"\"\") # Modality selection modality = gr.Radio( choices=["CT", "CR", "DX", "RX", "DR"], value="CT", label="Modality", info="Will be auto-detected for DICOM files" ) # Task selection task = gr.Dropdown( choices=[ ("🎯 Point Analysis", "analyze_point"), ("🔬 Fat Segmentation (CT only)", "segment_fat"), ("📊 Full Analysis", "full_analysis") ], value="full_analysis", label="Analysis Task" ) # ROI settings with gr.Accordion("🎯 Region of Interest (ROI)", open=True): roi_x = gr.Slider(0, 512, 256, label="X Position", step=1) roi_y = gr.Slider(0, 512, 256, label="Y Position", step=1) roi_radius = gr.Slider(5, 50, 10, label="Radius", step=1) # Clinical context with gr.Accordion("🏥 Clinical Context", open=False): symptoms = gr.CheckboxGroup( choices=[ "dyspnea", "chest_pain", "abdominal_pain", "trauma", "obesity_screening", "routine_check" ], label="Symptoms/Indication" ) # Visualization options with gr.Accordion("🎨 Visualization Options", open=True): show_overlay = gr.Checkbox( label="Show ROI/Segmentation Overlay", value=True, info="Display ROI circle or fat segmentation info on the image" ) analyze_btn = gr.Button("🔬 Analyze", variant="primary", size="lg") with gr.Column(scale=2): # Results with tabs for different views with gr.Tab("🖼️ Original Image"): image_display = gr.Image(label="Medical Image", type="numpy") with gr.Tab("🎯 Overlay View"): overlay_display = gr.Image(label="Image with Overlay", type="numpy") file_info = gr.Textbox(label="File Information", lines=1) with gr.Tab("📊 Visual Report"): report_html = gr.HTML() with gr.Tab("🔧 JSON Output"): json_output = gr.JSON(label="Structured Data for AI Agents") # Examples and help with gr.Row(): gr.Markdown(\"\"\" ### 📁 Supported Formats - **DICOM**: Automatic HU value extraction and modality detection - **PNG/JPG**: Interpreted based on selected modality - **All Formats**: Automatic grayscale conversion - **Files without extension**: Supported (e.g., IM_0001) - will try DICOM first ### 🎯 Usage 1. Upload a medical image file 2. Select modality (auto-detected for DICOM) 3. Choose analysis task 4. Adjust ROI position for point analysis 5. Click "Analyze" ### 💡 Features - **ROI Visualization**: See the exact area being analyzed - **Fat Segmentation**: Visual percentages for CT scans - **Multi-format Support**: Works with any medical image format - **AI Agent Ready**: Structured JSON output for integration \"\"\") # Connect the interface analyze_btn.click( fn=process_and_analyze, inputs=[file_input, modality, task, roi_x, roi_y, roi_radius, symptoms, show_overlay], outputs=[image_display, file_info, report_html, json_output, overlay_display] ) # Auto-update ROI limits when image is loaded def update_roi_on_upload(file_obj): if file_obj is None: return gr.update(), gr.update() try: analyzer = MedicalImageAnalyzer() _, _, metadata = analyzer.process_file(file_obj.name if hasattr(file_obj, 'name') else str(file_obj)) if 'shape' in metadata: h, w = metadata['shape'] return gr.update(maximum=w-1, value=w//2), gr.update(maximum=h-1, value=h//2) except: pass return gr.update(), gr.update() file_input.change( fn=update_roi_on_upload, inputs=[file_input], outputs=[roi_x, roi_y] ) return demo if __name__ == "__main__": demo = create_demo() demo.launch() ``` """, elem_classes=["md-custom"], header_links=True) gr.Markdown(""" ## `MedicalImageAnalyzer` ### Initialization """, elem_classes=["md-custom"], header_links=True) gr.ParamViewer(value=_docs["MedicalImageAnalyzer"]["members"]["__init__"], linkify=[]) gr.Markdown("### Events") gr.ParamViewer(value=_docs["MedicalImageAnalyzer"]["events"], linkify=['Event']) gr.Markdown(""" ### User function The impact on the users predict function varies depending on whether the component is used as an input or output for an event (or both). - When used as an Input, the component only impacts the input signature of the user function. - When used as an output, the component only impacts the return signature of the user function. The code snippet below is accurate in cases where the component is used as both an input and an output. ```python def predict( value: typing.Dict[str, typing.Any][str, typing.Any] ) -> typing.Dict[str, typing.Any][str, typing.Any]: return value ``` """, elem_classes=["md-custom", "MedicalImageAnalyzer-user-fn"], header_links=True) demo.load(None, js=r"""function() { const refs = {}; const user_fn_refs = { MedicalImageAnalyzer: [], }; requestAnimationFrame(() => { Object.entries(user_fn_refs).forEach(([key, refs]) => { if (refs.length > 0) { const el = document.querySelector(`.${key}-user-fn`); if (!el) return; refs.forEach(ref => { el.innerHTML = el.innerHTML.replace( new RegExp("\\b"+ref+"\\b", "g"), `${ref}` ); }) } }) Object.entries(refs).forEach(([key, refs]) => { if (refs.length > 0) { const el = document.querySelector(`.${key}`); if (!el) return; refs.forEach(ref => { el.innerHTML = el.innerHTML.replace( new RegExp("\\b"+ref+"\\b", "g"), `${ref}` ); }) } }) }) } """) demo.launch()