| import os |
| import time |
| import csv |
| import hashlib |
| from datetime import datetime |
| import numpy as np |
| import pydicom |
|
|
| import torch |
| import gradio as gr |
| from PIL import Image |
| from transformers import pipeline |
| from huggingface_hub import HfApi, hf_hub_download |
| from reportlab.pdfgen import canvas |
| from reportlab.lib.pagesizes import letter |
| from reportlab.lib.units import inch |
|
|
| MODEL_ID = "google/medgemma-1.5-4b-it" |
|
|
| PROMPT = ( |
| "You are a senior consultant radiologist. Write a detailed, structured chest X-ray report.\n\n" |
| "Use this exact format:\n" |
| "1) Technique: (projection PA/AP, penetration; if uncertain, say so)\n" |
| "2) Findings:\n" |
| " - Airways\n" |
| " - Lungs & pleura (consolidation, atelectasis, effusion, pneumothorax, interstitial markings, and any other relevant findings)\n" |
| " - Cardiomediastinal silhouette (heart size, mediastinum, hila)\n" |
| " - Diaphragm\n" |
| " - Bones & soft tissues\n" |
| " - Devices/lines (if any)\n" |
| "3) Impression: (bullet points, most likely diagnosis + key differentials)\n" |
| "4) Urgent alerts: (state 'None' if none)\n\n" |
| "Be specific about location and limitations. Do not invent clinical history. " |
| "If something is not clearly visible, say 'cannot be confidently assessed'." |
| ) |
|
|
| RATING_MAP = { |
| "1 - Very Unlikely": 1, |
| "2 - Unlikely": 2, |
| "3 - Neutral": 3, |
| "4 - Likely": 4, |
| "5 - Very Likely": 5, |
| } |
|
|
| LOCAL_TMP_DIR = "/tmp" |
| REVIEWS_CSV_NAME = "radiologist_reviews.csv" |
| LOCAL_REVIEWS_CSV = os.path.join(LOCAL_TMP_DIR, REVIEWS_CSV_NAME) |
|
|
| |
| hf_token = os.environ.get("HF_TOKEN") |
| if not hf_token: |
| raise RuntimeError("HF_TOKEN is missing. Add it in your Space secrets.") |
|
|
| dataset_repo = os.environ.get("HF_DATASET_REPO") |
| if not dataset_repo: |
| raise RuntimeError("HF_DATASET_REPO is missing. Add it in your Space variables.") |
|
|
| |
| api = HfApi(token=hf_token) |
|
|
| |
| use_cuda = torch.cuda.is_available() |
| dtype = torch.bfloat16 if use_cuda else torch.float32 |
|
|
| pipe = pipeline( |
| "image-text-to-text", |
| model=MODEL_ID, |
| dtype=dtype, |
| device=0 if use_cuda else -1, |
| token=hf_token, |
| ) |
|
|
| def generate_report(img: Image.Image): |
| if img is None: |
| return "Please upload a chest X-ray image.", None |
|
|
| img = img.convert("L").convert("RGB") |
| |
| messages = [{ |
| "role": "user", |
| "content": [ |
| {"type": "image", "image": img}, |
| {"type": "text", "text": PROMPT}, |
| ], |
| }] |
| |
| out = pipe( |
| text=messages, |
| max_new_tokens=500, |
| do_sample=False, |
| temperature=0.0, |
| top_p=1.0, |
| ) |
| |
| report = out[0]["generated_text"][-1]["content"] |
| return report, None |
|
|
| def text_to_pdf(report_text: str): |
| if not report_text or not report_text.strip(): |
| return None |
|
|
| filename = f"radiology_report_{int(time.time())}.pdf" |
| path = os.path.join("/tmp", filename) |
|
|
| c = canvas.Canvas(path, pagesize=letter) |
| width, height = letter |
|
|
| left = 0.75 * inch |
| top = height - 0.75 * inch |
| line_height = 14 |
| max_width = width - 2 * left |
|
|
| c.setFont("Helvetica-Bold", 14) |
| c.drawString(left, top, "Chest X-ray Report") |
| c.setFont("Helvetica", 11) |
|
|
| y = top - 0.45 * inch |
|
|
| for paragraph in report_text.split("\n"): |
| words = paragraph.split(" ") |
| line = "" |
|
|
| for w in words: |
| test = (line + " " + w).strip() |
| if c.stringWidth(test, "Helvetica", 11) <= max_width: |
| line = test |
| else: |
| if y < 0.75 * inch: |
| c.showPage() |
| c.setFont("Helvetica", 11) |
| y = height - 0.75 * inch |
| if line: |
| c.drawString(left, y, line) |
| y -= line_height |
| line = w |
|
|
| if y < 0.75 * inch: |
| c.showPage() |
| c.setFont("Helvetica", 11) |
| y = height - 0.75 * inch |
|
|
| if line: |
| c.drawString(left, y, line) |
| y -= line_height |
|
|
| c.save() |
| return path |
|
|
| def dicom_to_png(dicom_file): |
| if dicom_file is None: |
| return None, "Please upload a DICOM file." |
|
|
| try: |
| ds = pydicom.dcmread(dicom_file.name) |
| pixel_array = ds.pixel_array.astype(float) |
|
|
| |
| pixel_min = pixel_array.min() |
| pixel_max = pixel_array.max() |
| if pixel_max - pixel_min == 0: |
| return None, "Image has no contrast — cannot convert." |
|
|
| normalized = (pixel_array - pixel_min) / (pixel_max - pixel_min) * 255.0 |
| img_uint8 = normalized.astype(np.uint8) |
|
|
| |
| if img_uint8.ndim == 2: |
| img = Image.fromarray(img_uint8, mode="L").convert("RGB") |
| elif img_uint8.ndim == 3: |
| img = Image.fromarray(img_uint8) |
| else: |
| return None, "Unsupported pixel array dimensions." |
|
|
| |
| out_path = os.path.join(LOCAL_TMP_DIR, f"converted_{int(time.time())}.png") |
| img.save(out_path, format="PNG") |
|
|
| return out_path, "Conversion successful. Download your PNG below." |
|
|
| except Exception as e: |
| return None, f"Conversion failed: {str(e)}" |
|
|
| def load_image_from_upload(upload) -> tuple[Image.Image | None, str]: |
| """ |
| Accepts either a PIL Image (from gr.Image) or a file object (from gr.File). |
| If DICOM, converts internally. Returns (PIL Image, error_message). |
| """ |
| if upload is None: |
| return None, "Please upload a chest X-ray image." |
|
|
| |
| if isinstance(upload, Image.Image): |
| return upload, "" |
|
|
| |
| filepath = upload.name if hasattr(upload, "name") else str(upload) |
| ext = os.path.splitext(filepath)[-1].lower() |
|
|
| if ext == ".dcm": |
| try: |
| ds = pydicom.dcmread(filepath) |
| pixel_array = ds.pixel_array.astype(float) |
| pixel_min, pixel_max = pixel_array.min(), pixel_array.max() |
|
|
| if pixel_max - pixel_min == 0: |
| return None, "DICOM image has no contrast — cannot convert." |
|
|
| normalized = (pixel_array - pixel_min) / (pixel_max - pixel_min) * 255.0 |
| img_uint8 = normalized.astype(np.uint8) |
|
|
| if img_uint8.ndim == 2: |
| img = Image.fromarray(img_uint8, mode="L").convert("RGB") |
| elif img_uint8.ndim == 3: |
| img = Image.fromarray(img_uint8) |
| else: |
| return None, "Unsupported DICOM pixel format." |
|
|
| return img, "" |
| except Exception as e: |
| return None, f"DICOM conversion failed: {str(e)}" |
|
|
| else: |
| |
| try: |
| img = Image.open(filepath).convert("RGB") |
| return img, "" |
| except Exception as e: |
| return None, f"Could not open image: {str(e)}" |
|
|
| def download_existing_csv() -> str: |
| os.makedirs(LOCAL_TMP_DIR, exist_ok=True) |
|
|
| try: |
| csv_path = hf_hub_download( |
| repo_id=dataset_repo, |
| repo_type="dataset", |
| filename=REVIEWS_CSV_NAME, |
| token=hf_token, |
| local_dir=LOCAL_TMP_DIR, |
| local_dir_use_symlinks=False, |
| ) |
| return csv_path |
| except Exception: |
| with open(LOCAL_REVIEWS_CSV, "w", newline="", encoding="utf-8") as f: |
| writer = csv.writer(f) |
| writer.writerow([ |
| "Review_ID", |
| "Timestamp_utc", |
| "Report_hash", |
| "Technique_label", "Technique_value", |
| "Findings_label", "Findings_value", |
| "Impression_label","Impression_value", |
| "Image_filename", |
| ]) |
| return LOCAL_REVIEWS_CSV |
|
|
| def upload_csv_to_dataset(csv_path: str): |
| api.upload_file( |
| path_or_fileobj=csv_path, |
| path_in_repo=REVIEWS_CSV_NAME, |
| repo_id=dataset_repo, |
| repo_type="dataset", |
| commit_message="Update radiologist review ratings CSV", |
| ) |
|
|
| def get_next_review_id(csv_path: str) -> int: |
| """Count existing data rows and return next ID.""" |
| with open(csv_path, "r", encoding="utf-8") as f: |
| reader = csv.reader(f) |
| rows = list(reader) |
| |
| return len(rows) |
|
|
|
|
| def get_report_hash(report_text: str) -> str: |
| """MD5 hash of the report text to detect duplicates.""" |
| return hashlib.md5(report_text.strip().encode("utf-8")).hexdigest() |
|
|
|
|
| def is_duplicate_review(csv_path: str, report_hash: str) -> bool: |
| """Check if this report hash already exists in the CSV.""" |
| with open(csv_path, "r", encoding="utf-8") as f: |
| reader = csv.DictReader(f) |
| for row in reader: |
| if row.get("Report_hash") == report_hash: |
| return True |
| return False |
|
|
|
|
| def upload_image_to_dataset(img: Image.Image, review_id: int) -> str: |
| """Save image locally and upload to dataset repo. Returns filename.""" |
| filename = f"review_{review_id}.jpg" |
| local_path = os.path.join(LOCAL_TMP_DIR, filename) |
| img.convert("RGB").save(local_path, format="JPEG") |
|
|
| api.upload_file( |
| path_or_fileobj=local_path, |
| path_in_repo=f"images/{filename}", |
| repo_id=dataset_repo, |
| repo_type="dataset", |
| commit_message=f"Upload image for review {review_id}", |
| ) |
| return filename |
|
|
| def save_review(report_text: str, technique_label: str, findings_label: str, impression_label: str, img: Image.Image = None): |
| if not report_text or not report_text.strip(): |
| return "No report available to review." |
|
|
| if not technique_label or not findings_label or not impression_label: |
| return "Please select a rating for all three sections before saving." |
|
|
| try: |
| csv_path = download_existing_csv() |
|
|
| |
| report_hash = get_report_hash(report_text) |
| if is_duplicate_review(csv_path, report_hash): |
| return "This report has already been reviewed. Duplicate not saved." |
|
|
| |
| review_id = get_next_review_id(csv_path) |
|
|
| |
| image_filename = "N/A" |
| if img is not None: |
| image_filename = upload_image_to_dataset(img, review_id) |
|
|
| with open(csv_path, "a", newline="", encoding="utf-8") as f: |
| writer = csv.writer(f) |
| writer.writerow([ |
| review_id, |
| datetime.utcnow().isoformat(), |
| report_hash, |
| technique_label, RATING_MAP[technique_label], |
| findings_label, RATING_MAP[findings_label], |
| impression_label, RATING_MAP[impression_label], |
| image_filename, |
| ]) |
|
|
| upload_csv_to_dataset(csv_path) |
| return f"Review #{review_id} saved successfully to dataset repo: {dataset_repo}" |
|
|
| except Exception as e: |
| return f"Failed to save review: {str(e)}" |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| DARK_CSS = """ |
| /* ── Base ── */ |
| body, .gradio-container { |
| background: #0a1628 !important; |
| color: #e8edf5 !important; |
| font-family: 'Segoe UI', system-ui, -apple-system, sans-serif !important; |
| } |
| |
| /* Centre & constrain */ |
| .gradio-container { |
| max-width: 920px !important; |
| margin-left: auto !important; |
| margin-right: auto !important; |
| width: 100% !important; |
| } |
| |
| /* ── Headings & body text ── */ |
| .gr-markdown, |
| .gr-markdown h1, .gr-markdown h2, .gr-markdown h3, |
| .gr-markdown p, h1, h2, h3, p { |
| color: #e8edf5 !important; |
| } |
| |
| h1 { |
| font-size: 1.75rem !important; |
| font-weight: 700 !important; |
| letter-spacing: -0.3px !important; |
| color: #ffffff !important; |
| } |
| |
| /* ── Labels ── */ |
| label, .gr-label, .wrap label { |
| color: #93aac9 !important; |
| font-size: 0.8rem !important; |
| font-weight: 600 !important; |
| letter-spacing: 0.06em !important; |
| text-transform: uppercase !important; |
| } |
| |
| /* ── Panels / containers ── */ |
| .gr-panel, .gr-box, .gr-form { |
| background: #112040 !important; |
| border: 1px solid #1e3560 !important; |
| border-radius: 12px !important; |
| } |
| |
| /* ── Textbox — report text stays near-black on white ── */ |
| textarea, .wrap textarea { |
| background: #ffffff !important; |
| color: #111827 !important; |
| border: 1px solid #1e3560 !important; |
| border-radius: 10px !important; |
| font-size: 0.9rem !important; |
| line-height: 1.7 !important; |
| } |
| |
| textarea:focus, .wrap textarea:focus { |
| border-color: #4a90d9 !important; |
| box-shadow: 0 0 0 3px rgba(74, 144, 217, 0.2) !important; |
| outline: none !important; |
| } |
| |
| /* ── Buttons — base ── */ |
| button { |
| background: transparent !important; |
| color: #e8edf5 !important; |
| border: 1.5px solid #2e5080 !important; |
| border-radius: 8px !important; |
| font-weight: 600 !important; |
| font-size: 0.875rem !important; |
| letter-spacing: 0.02em !important; |
| padding: 10px 20px !important; |
| transition: all 0.18s ease !important; |
| } |
| |
| /* ── Primary action (Generate) — accent fill ── */ |
| button.primary, .gr-button-primary, #component-gen-btn { |
| background: #1a6de0 !important; |
| color: #ffffff !important; |
| border-color: #1a6de0 !important; |
| box-shadow: 0 2px 12px rgba(26, 109, 224, 0.35) !important; |
| } |
| |
| button.primary:hover:not(:disabled), |
| .gr-button-primary:hover:not(:disabled), |
| #component-gen-btn:hover:not(:disabled) { |
| background: #1558c0 !important; |
| border-color: #1558c0 !important; |
| box-shadow: 0 4px 18px rgba(26, 109, 224, 0.5) !important; |
| transform: translateY(-1px) !important; |
| } |
| |
| /* ── All other button hovers ── */ |
| button:hover:not(:disabled) { |
| background: #1e3560 !important; |
| border-color: #4a90d9 !important; |
| color: #ffffff !important; |
| transform: translateY(-1px) !important; |
| box-shadow: 0 3px 10px rgba(0,0,0,0.25) !important; |
| } |
| |
| button:active:not(:disabled) { |
| transform: translateY(0) !important; |
| box-shadow: none !important; |
| } |
| |
| /* ── Disabled state ── */ |
| button:disabled { |
| opacity: 0.3 !important; |
| cursor: not-allowed !important; |
| transform: none !important; |
| box-shadow: none !important; |
| } |
| |
| /* ── Radio buttons ── */ |
| .gr-radio, .wrap .gr-radio { |
| color: #e8edf5 !important; |
| } |
| |
| input[type="radio"]:checked + span { |
| color: #4a90d9 !important; |
| font-weight: 600 !important; |
| } |
| |
| /* ── File upload zone ── */ |
| input[type="file"] { |
| color: #93aac9 !important; |
| } |
| |
| .gradio-file-upload, [data-testid="file-drop-area"] { |
| background: #0d1f3c !important; |
| border: 1.5px dashed #2e5080 !important; |
| border-radius: 12px !important; |
| transition: border-color 0.2s, background 0.2s !important; |
| } |
| |
| .gradio-file-upload:hover, [data-testid="file-drop-area"]:hover { |
| border-color: #4a90d9 !important; |
| background: #112040 !important; |
| } |
| |
| .gradio-file-upload svg { |
| fill: #4a90d9 !important; |
| stroke: #4a90d9 !important; |
| } |
| |
| /* ── Image preview ── */ |
| .gr-image, [data-testid="image"] { |
| border: 1px solid #1e3560 !important; |
| border-radius: 12px !important; |
| overflow: hidden !important; |
| } |
| |
| /* ── Scrollbar ── */ |
| ::-webkit-scrollbar { width: 5px; height: 5px; } |
| ::-webkit-scrollbar-t |
| """ |
|
|
| with gr.Blocks() as demo: |
| gr.Markdown("# Chest X-ray Auto Report Generator\n Radiologist Assistant") |
| |
| image_in = gr.File( |
| label="Upload chest X-ray (PNG / JPG / DICOM .dcm)", |
| file_types=[".png", ".jpg", ".jpeg", ".dcm"] |
| ) |
|
|
| image_preview = gr.Image( |
| label="Image Preview", |
| type="pil", |
| interactive=False, |
| height=400, |
| visible = False, |
| ) |
| |
| report_loading = gr.HTML( |
| """ |
| <div style=" |
| height: 420px; |
| border: 1px solid #d1d5db; |
| border-radius: 8px; |
| background: black; |
| display: flex; |
| align-items: center; |
| justify-content: center; |
| flex-direction: column; |
| gap: 10px; |
| padding: 16px; |
| "> |
| <img src="https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExYW5pdDh0ZHdoNXplaTlqdno1OWNrbzc3cDcxaTRkc3FoaTN1anF2aiZlcD12MV9naWZzX3NlYXJjaCZjdD1n/dt0KXLj7bzwZuRQBwY/giphy.gif" |
| style="width:200px; height:auto;" /> |
| <div style="font-size:25px; font-weight:800; color:white;">Generating...</div> |
| </div> |
| """, |
| visible=False, |
| ) |
| |
| report_out = gr.Textbox( |
| label="Generated radiology report", |
| lines=18, |
| interactive=False, |
| visible=True, |
| ) |
| |
| review_done = gr.State(False) |
| |
| review_section = gr.Column(visible=False) |
| |
| LIKERT_CHOICES = [ |
| "1 - Very Unlikely", |
| "2 - Unlikely", |
| "3 - Neutral", |
| "4 - Likely", |
| "5 - Very Likely", |
| ] |
| |
| with review_section: |
| gr.Markdown("Radiologist Review") |
| gr.Markdown("Rate each section of the generated report:") |
| |
| technique_radio = gr.Radio( |
| choices=LIKERT_CHOICES, |
| label="1. Technique — accuracy & completeness", |
| value=None |
| ) |
| findings_radio = gr.Radio( |
| choices=LIKERT_CHOICES, |
| label="2. Findings — accuracy & completeness", |
| value=None |
| ) |
| impression_radio = gr.Radio( |
| choices=LIKERT_CHOICES, |
| label="3. Impression — accuracy & completeness", |
| value=None |
| ) |
| |
| save_review_btn = gr.Button("Save Review") |
| review_status = gr.Textbox(label="Review Status", interactive=False) |
|
|
| def preview_only(upload): |
| img, error = load_image_from_upload(upload) |
| if img is None: |
| return gr.update(value=None, visible=False), gr.update(interactive=False) |
|
|
| return gr.update(value=img, visible=True), gr.update(interactive=True) |
|
|
| def clear_all(): |
| return ( |
| gr.update(value=None, visible=False), |
| gr.update(value="", visible=True), |
| gr.update(visible=False), |
| gr.update(visible=False), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=""), |
| False, |
| ) |
| |
| gen_btn = gr.Button("Generate report", interactive=False) |
| pdf_btn = gr.DownloadButton("Download as PDF", value=None, visible=True) |
| |
| def generate_report_ui(upload): |
| img, error = load_image_from_upload(upload) |
| |
| if img is None: |
| yield ( |
| gr.update(value=None), |
| gr.update(value="error", interactive=False, visible=True), |
| gr.update(visible=False), |
| gr.update(visible=False), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=""), |
| False, |
| ) |
| return |
| |
| |
| yield ( |
| gr.update(value=img, visible=True), |
| gr.update(value="", interactive=False, visible=False), |
| gr.update(visible=True), |
| gr.update(visible=False), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=""), |
| False, |
| ) |
| |
| report, _ = generate_report(img) |
| |
| |
| yield ( |
| gr.update(value=img, visible=True), |
| gr.update(value=report, interactive=True, visible=True), |
| gr.update(visible=False), |
| gr.update(visible=True), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=None), |
| gr.update(value=""), |
| False, |
| ) |
| |
| gen_btn.click( |
| fn=generate_report_ui, |
| inputs=image_in, |
| outputs=[ |
| image_preview, |
| report_out, |
| report_loading, |
| review_section, |
| technique_radio, |
| findings_radio, |
| impression_radio, |
| review_status, |
| review_done, |
| ], |
| ) |
| |
| def guarded_pdf_download(report_text, review_done_flag): |
| if not review_done_flag: |
| gr.Warning("Please submit a review before downloading the report.") |
| return None |
| return text_to_pdf(report_text) |
| |
| pdf_btn.click( |
| fn=guarded_pdf_download, |
| inputs=[report_out, review_done], |
| outputs=pdf_btn, |
| ) |
| |
| def save_review_ui(report_text, technique, findings, impression, upload): |
| img, _ = load_image_from_upload(upload) |
| message = save_review(report_text, technique, findings, impression, img) |
| |
| if "successfully" in message.lower(): |
| return message, True |
| else: |
| return message, False |
| |
| save_review_btn.click( |
| fn=save_review_ui, |
| inputs=[report_out, technique_radio, findings_radio, impression_radio, image_in], |
| outputs=[review_status, review_done], |
| ) |
|
|
| image_in.change( |
| fn=preview_only, |
| inputs=image_in, |
| outputs=[image_preview, gen_btn], |
| ) |
|
|
| |
| image_in.clear( |
| fn=clear_all, |
| outputs=[ |
| image_preview, |
| report_out, |
| report_loading, |
| review_section, |
| technique_radio, |
| findings_radio, |
| impression_radio, |
| review_status, |
| review_done, |
| ] |
| ) |
|
|
| demo.launch( |
| theme=gr.themes.Monochrome(), |
| css=DARK_CSS, |
| ) |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
|
|
| |
| |
| |
|
|
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
|
|
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
| |
| |
|
|
