Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import base64 | |
| import re | |
| import os | |
| import tempfile | |
| from openai import OpenAI | |
| from sarvamai import SarvamAI | |
| from sarvamai.play import save | |
| # ─── API KEYS ───────────────────────────────────────────────────────────────── | |
| OPENAI_KEY = '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' | |
| SARVAM_KEY = 'sk_lnnt2cq6_BsU3lYlvinKhKfR4x2v8WqGC' | |
| # ─── SYSTEM PROMPT ──────────────────────────────────────────────────────────── | |
| SYSTEM_PROMPT = """ | |
| You are a medical report summarization assistant. | |
| STRICT RULES: | |
| 1. Only process medical report images (lab reports, prescriptions, scans, test reports). | |
| 2. If the image is NOT a medical report, respond ONLY with: | |
| "ERROR: The uploaded image is not a valid medical report." | |
| 3. Summarize in VERY SIMPLE language so a non-medical person can understand. | |
| 4. Avoid medical jargon. If needed, explain it simply. | |
| Your response MUST follow this EXACT structure with these two section headers: | |
| ## DETAILED SUMMARY | |
| Write a thorough, easy-to-understand breakdown of the report. | |
| Cover: key values, abnormal findings (in simple terms), what each result means for | |
| the person's health, and anything they should be aware of. | |
| End this section with: "AI Generated Summary: Always consult a doctor before taking any further steps." | |
| ## CONCISE SUMMARY | |
| Write a short paragraph of AT LEAST 40 WORDS and NO MORE THAN 50 WORDS. | |
| It must cover ALL key findings — normal results, abnormal values, and what the person | |
| should watch out for. Miss nothing important. Every aspect of the report must be reflected. | |
| """ | |
| # ─── HELPERS ────────────────────────────────────────────────────────────────── | |
| def encode_image(path: str) -> str: | |
| with open(path, "rb") as f: | |
| return base64.b64encode(f.read()).decode("utf-8") | |
| def extract_section(text: str, header: str) -> str: | |
| pattern = rf"##\s*{re.escape(header)}\s*\n(.*?)(?=\n##\s|\Z)" | |
| match = re.search(pattern, text, re.DOTALL | re.IGNORECASE) | |
| return match.group(1).strip() if match else "" | |
| def generate_tts(text: str, api_key: str) -> tuple[str | None, str]: | |
| """Generate TTS audio using Sarvam AI and return (path_to_wav, status_message).""" | |
| try: | |
| if not text or not text.strip(): | |
| return None, "No text provided for TTS" | |
| client = SarvamAI(api_subscription_key=api_key) | |
| audio = client.text_to_speech.convert( | |
| target_language_code="en-IN", | |
| text=text.strip(), | |
| model="bulbul:v3", | |
| speaker="varun", | |
| ) | |
| tmp = tempfile.NamedTemporaryFile( | |
| delete=False, suffix=".wav", dir=tempfile.gettempdir() | |
| ) | |
| tmp.close() | |
| save(audio, tmp.name) | |
| if not os.path.exists(tmp.name) or os.path.getsize(tmp.name) == 0: | |
| return None, "TTS generated empty audio" | |
| return tmp.name, "TTS generated successfully" | |
| except Exception as e: | |
| return None, f"[TTS ERROR] {str(e)}" | |
| # ─── CORE LOGIC ─────────────────────────────────────────────────────────────── | |
| def analyse_report(image_path): | |
| if image_path is None: | |
| return ( | |
| "<div class='error-box'>⚠️ PLEASE UPLOAD A MEDICAL DOCUMENT FIRST.</div>", | |
| "", None, "" | |
| ) | |
| openai_client = OpenAI(api_key=OPENAI_KEY) | |
| img_b64 = encode_image(image_path) | |
| # ── GPT-4o call ────────────────────────────────────────────────────────── | |
| full_output = "" | |
| try: | |
| with openai_client.responses.stream( | |
| model="gpt-4o", | |
| input=[ | |
| {"role": "system", "content": SYSTEM_PROMPT}, | |
| { | |
| "role": "user", | |
| "content": [ | |
| {"type": "input_text", "text": "Analyze and summarize this medical report"}, | |
| {"type": "input_image", "image_url": f"data:image/jpeg;base64,{img_b64}"} | |
| ] | |
| } | |
| ] | |
| ) as stream: | |
| for event in stream: | |
| if event.type == "response.output_text.delta": | |
| full_output += event.delta | |
| final_response = stream.get_final_response() | |
| except Exception as e: | |
| return (f"<div class='error-box'>❌ OPENAI ERROR: {e}</div>", "", None, "") | |
| if full_output.strip().startswith("ERROR:"): | |
| return ( | |
| f"<div class='error-box'>⚠️ {full_output.strip()}</div>", | |
| "", None, "" | |
| ) | |
| detailed_summary = extract_section(full_output, "DETAILED SUMMARY") | |
| summary_50 = extract_section(full_output, "CONCISE SUMMARY") | |
| words = summary_50.split() | |
| if len(words) > 50: | |
| summary_50 = " ".join(words[:50]) + "..." | |
| # ── Token cost ──────────────────────────────────────────────────────────── | |
| usage = final_response.usage | |
| total_cost = (usage.input_tokens * 2.5 + usage.output_tokens * 10.0) / 1_000_000 | |
| metrics_html = f""" | |
| <div class='metrics-row'> | |
| <div class='chip'><span class='val'>{usage.input_tokens:,}</span><span class='lbl'>INPUT TOKENS</span></div> | |
| <div class='chip'><span class='val'>{usage.output_tokens:,}</span><span class='lbl'>OUTPUT TOKENS</span></div> | |
| <div class='chip'><span class='val'>${total_cost:.5f}</span><span class='lbl'>ANALYSIS COST</span></div> | |
| </div>""" | |
| # ── Sarvam AI TTS ───────────────────────────────────────────────────────── | |
| audio_path, tts_status = generate_tts(summary_50, SARVAM_KEY) | |
| safe_detailed = detailed_summary.replace("<", "<").replace(">", ">") | |
| safe_concise = summary_50.replace("<", "<").replace(">", ">") | |
| detailed_html = f""" | |
| <div class='result-card detailed'> | |
| <div class='card-header'> | |
| <div class='card-icon blue'>❖</div> | |
| <div class='card-title'>COMPREHENSIVE BREAKDOWN</div> | |
| </div> | |
| <div class='card-body'>{safe_detailed}</div> | |
| {metrics_html} | |
| </div>""" | |
| # Animated Waveform integration | |
| waveform_html = """ | |
| <div class="waveform-container"> | |
| <div class="bar"></div><div class="bar"></div><div class="bar"></div><div class="bar"></div><div class="bar"></div> | |
| </div> | |
| """ if audio_path else "" | |
| tts_indicator = ( | |
| f'<div class="audio-badge ready">{waveform_html} AUDIO SIGNAL ACTIVE</div>' | |
| if audio_path else | |
| '<div class="audio-badge failed">✗ AUDIO SIGNAL OFFLINE</div>' | |
| ) | |
| concise_html = f""" | |
| <div class='result-card concise'> | |
| <div class='card-header' style="justify-content: space-between;"> | |
| <div style="display:flex; align-items:center; gap:10px;"> | |
| <div class='card-icon coral'>⚡</div> | |
| <div class='card-title'>QUICK BRIEFING</div> | |
| </div> | |
| {tts_indicator} | |
| </div> | |
| <div class='card-body focus-text'>{safe_concise}</div> | |
| </div>""" | |
| disclaimer = """ | |
| <div class='disclaimer'> | |
| <span style="font-size:18px;">⚠️</span> | |
| <span><strong>MEDICAL OVERRIDE WARNING:</strong> THIS INTELLIGENCE IS AI-GENERATED FOR INFORMATIONAL PURPOSES. IT DOES NOT REPLACE PROFESSIONAL MEDICAL ADVICE. ALWAYS CONSULT YOUR HEALTHCARE PROVIDER.</span> | |
| </div>""" | |
| return detailed_html, concise_html, audio_path, disclaimer | |
| # ─── CUSTOM CSS (Neon Cyberpunk UI/UX) ───────────────────────────────────────── | |
| CSS = """ | |
| @import url('https://fonts.googleapis.com/css2?family=Outfit:wght@300;400;500;700&family=Share+Tech+Mono&display=swap'); | |
| :root { | |
| --bg-main: #000000; | |
| --bg-card: #080808; | |
| --primary: #00FFFF; /* Neon Cyan */ | |
| --primary-glow: rgba(0, 255, 255, 0.6); | |
| --primary-soft: rgba(0, 255, 255, 0.1); | |
| --accent: #FF00FF; /* Neon Magenta */ | |
| --accent-glow: rgba(255, 0, 255, 0.6); | |
| --accent-soft: rgba(255, 0, 255, 0.1); | |
| --border-soft: rgba(0, 255, 255, 0.3); | |
| --text-neon: #E0FFFF; /* Bright icy blue for main readability */ | |
| --text-muted: #008B8B; | |
| --shadow-glow: 0 0 10px var(--primary-glow), inset 0 0 10px var(--primary-soft); | |
| --shadow-accent: 0 0 10px var(--accent-glow), inset 0 0 10px var(--accent-soft); | |
| } | |
| body, .gradio-container { | |
| background: var(--bg-main) !important; | |
| font-family: 'Share Tech Mono', monospace !important; | |
| color: var(--text-neon) !important; | |
| } | |
| .gradio-container { max-width: 1280px !important; margin: 0 auto !important; padding: 20px !important; } | |
| footer { display: none !important; } | |
| /* HERO SECTION */ | |
| .hero { text-align: left; padding: 20px 10px 40px; } | |
| .hero-title { | |
| font-family: 'Outfit', sans-serif; | |
| font-size: clamp(32px, 4vw, 48px); font-weight: 700; | |
| color: #FFF; margin: 0; letter-spacing: 2px; text-transform: uppercase; | |
| text-shadow: 0 0 8px var(--primary), 0 0 20px var(--primary); | |
| } | |
| .hero-title span { | |
| color: #FFF; | |
| text-shadow: 0 0 8px var(--accent), 0 0 20px var(--accent); | |
| } | |
| .hero-sub { | |
| font-family: 'Share Tech Mono', monospace; font-size: 16px; | |
| color: var(--primary); margin: 12px 0 0; line-height: 1.6; max-width: 550px; | |
| text-shadow: 0 0 4px var(--primary); | |
| } | |
| /* PANELS (For single screen fitting) */ | |
| .left-panel { display: flex; flex-direction: column; gap: 20px; } | |
| .right-panel { | |
| display: flex; flex-direction: column; gap: 20px; | |
| max-height: 85vh; overflow-y: auto; padding-right: 15px; | |
| } | |
| .right-panel::-webkit-scrollbar { width: 4px; } | |
| .right-panel::-webkit-scrollbar-thumb { background: var(--primary); box-shadow: 0 0 5px var(--primary); border-radius: 10px; } | |
| /* UPLOAD AREA */ | |
| .upload-panel { | |
| background: var(--bg-card); border: 1px dashed var(--primary); | |
| border-radius: 12px; overflow: hidden; transition: all 0.3s ease; | |
| box-shadow: var(--shadow-glow); | |
| } | |
| .upload-panel:hover { border: 1px solid var(--accent); box-shadow: var(--shadow-accent); } | |
| .upload-panel .wrap { background: transparent !important; border: none !important; } | |
| .upload-panel svg { color: var(--primary) !important; filter: drop-shadow(0 0 5px var(--primary)); } | |
| .upload-panel span { color: var(--text-neon) !important; text-shadow: 0 0 5px var(--primary); } | |
| /* ACTION BUTTON */ | |
| #analyse-btn { | |
| background: var(--bg-card) !important; | |
| border: 1px solid var(--primary) !important; | |
| border-radius: 8px !important; color: var(--primary) !important; | |
| font-family: 'Share Tech Mono', monospace !important; font-size: 18px !important; | |
| letter-spacing: 2px !important; padding: 18px !important; | |
| box-shadow: var(--shadow-glow) !important; | |
| text-shadow: 0 0 5px var(--primary) !important; | |
| transition: all 0.2s ease !important; | |
| text-transform: uppercase; | |
| } | |
| #analyse-btn:hover { | |
| background: var(--primary-soft) !important; | |
| border-color: var(--accent) !important; | |
| color: var(--accent) !important; | |
| box-shadow: var(--shadow-accent) !important; | |
| text-shadow: 0 0 8px var(--accent) !important; | |
| transform: scale(1.02) !important; | |
| } | |
| /* RESULT CARDS */ | |
| .result-card { | |
| background: var(--bg-card); border: 1px solid var(--border-soft); | |
| border-radius: 12px; padding: 28px 32px; | |
| box-shadow: inset 0 0 15px rgba(0,255,255,0.05); position: relative; | |
| } | |
| .result-card::before { | |
| content: ''; position: absolute; top: 0; left: 0; width: 100%; height: 2px; | |
| background: linear-gradient(90deg, transparent, var(--primary), transparent); | |
| } | |
| .result-card.concise::before { | |
| background: linear-gradient(90deg, transparent, var(--accent), transparent); | |
| } | |
| .card-header { display: flex; align-items: center; gap: 12px; margin-bottom: 16px; border-bottom: 1px solid var(--border-soft); padding-bottom: 16px;} | |
| .card-icon { width: 36px; height: 36px; border-radius: 8px; display: flex; align-items: center; justify-content: center; font-size: 18px; border: 1px solid; } | |
| .card-icon.blue { background: var(--primary-soft); color: var(--primary); border-color: var(--primary); box-shadow: 0 0 8px var(--primary); text-shadow: 0 0 5px var(--primary); } | |
| .card-icon.coral { background: var(--accent-soft); color: var(--accent); border-color: var(--accent); box-shadow: 0 0 8px var(--accent); text-shadow: 0 0 5px var(--accent); } | |
| .card-title { font-size: 16px; font-weight: 700; letter-spacing: 0.1em; color: var(--primary); text-shadow: 0 0 5px var(--primary); } | |
| .result-card.concise .card-title { color: var(--accent); text-shadow: 0 0 5px var(--accent); } | |
| .card-body { font-family: 'Outfit', sans-serif; font-size: 16px; font-weight: 300; line-height: 1.8; color: var(--text-neon); white-space: pre-wrap; } | |
| .focus-text { font-size: 18px; font-weight: 400; line-height: 1.6; color: #FFF; text-shadow: 0 0 4px var(--accent); } | |
| /* METRICS */ | |
| .metrics-row { display: flex; gap: 12px; margin-top: 24px; } | |
| .chip { flex: 1; background: #000; border-radius: 8px; padding: 12px 14px; text-align: center; border: 1px solid var(--border-soft); box-shadow: inset 0 0 8px rgba(0,255,255,0.1); } | |
| .chip .val { display: block; font-size: 20px; font-weight: 700; color: var(--primary); text-shadow: 0 0 5px var(--primary); } | |
| .chip .lbl { display: block; font-size: 12px; color: var(--text-muted); margin-top: 4px; } | |
| /* AUDIO COMPONENT STYLING */ | |
| #audio-out { background: transparent !important; border: none !important; padding: 0 !important; margin-bottom: 10px !important; } | |
| #audio-out audio { | |
| width: 100% !important; border-radius: 8px !important; height: 50px !important; | |
| border: 1px solid var(--accent) !important; | |
| box-shadow: 0 0 10px var(--accent-glow) !important; | |
| filter: invert(1) hue-rotate(180deg); /* Forces a dark look on default audio players */ | |
| } | |
| /* WAVEFORM ANIMATION */ | |
| .audio-badge { display: flex; align-items: center; gap: 8px; font-size: 12px; font-weight: 700; padding: 6px 12px; border-radius: 4px; border: 1px solid; } | |
| .audio-badge.ready { background: var(--accent-soft); color: var(--accent); border-color: var(--accent); box-shadow: 0 0 8px var(--accent); text-shadow: 0 0 4px var(--accent); } | |
| .audio-badge.failed { background: rgba(255,0,0,0.1); color: #FF0000; border-color: #FF0000; box-shadow: 0 0 8px #FF0000; text-shadow: 0 0 4px #FF0000; } | |
| .waveform-container { display: flex; align-items: center; gap: 2px; height: 14px; margin-right: 4px; } | |
| .bar { width: 3px; background: var(--accent); border-radius: 1px; box-shadow: 0 0 5px var(--accent); animation: wave 1s ease-in-out infinite; } | |
| .bar:nth-child(1) { height: 40%; animation-delay: 0.0s; } | |
| .bar:nth-child(2) { height: 80%; animation-delay: 0.1s; } | |
| .bar:nth-child(3) { height: 100%; animation-delay: 0.2s; } | |
| .bar:nth-child(4) { height: 60%; animation-delay: 0.3s; } | |
| .bar:nth-child(5) { height: 40%; animation-delay: 0.4s; } | |
| @keyframes wave { 0%, 100% { transform: scaleY(0.4); } 50% { transform: scaleY(1); } } | |
| /* DISCLAIMER */ | |
| .disclaimer { | |
| background: rgba(255, 255, 0, 0.05); border-left: 4px solid #FFFF00; border-radius: 0 8px 8px 0; | |
| padding: 16px 20px; font-size: 13px; color: #FFFF00; display: flex; align-items: center; gap: 12px; line-height: 1.5; | |
| box-shadow: 0 0 10px rgba(255, 255, 0, 0.2); text-shadow: 0 0 4px rgba(255, 255, 0, 0.5); | |
| } | |
| .error-box { | |
| background: rgba(255,0,0,0.1); border: 1px solid #FF0000; padding: 15px; border-radius: 8px; | |
| color: #FF0000; text-shadow: 0 0 5px #FF0000; box-shadow: 0 0 10px rgba(255,0,0,0.3); text-align: center; | |
| } | |
| /* MOBILE RESPONSIVENESS */ | |
| @media (max-width: 900px) { | |
| .right-panel { max-height: none; overflow-y: visible; padding-right: 0; } | |
| .hero-title { font-size: 28px; } | |
| .card-header { flex-direction: column; align-items: flex-start; } | |
| } | |
| """ | |
| # ─── GRADIO UI ──────────────────────────────────────────────────────────────── | |
| with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="MediScan AI") as demo: | |
| with gr.Row(): | |
| # LEFT COLUMN (Uploader & Controls) | |
| with gr.Column(scale=4, elem_classes="left-panel"): | |
| gr.HTML(""" | |
| <div class="hero"> | |
| <div class="hero-title">MEDI<span>SCAN</span> AI</div> | |
| <p class="hero-sub">INITIALIZE UPLOAD. AI CLINICAL ENGINE STANDING BY FOR PLAIN-LANGUAGE DECRYPTION & AUDIO SYNTHESIS.</p> | |
| </div> | |
| """) | |
| with gr.Column(elem_classes="upload-panel"): | |
| image_input = gr.Image( | |
| type="filepath", | |
| label="DRAG & DROP MEDICAL DATAPAD HERE", | |
| show_label=True, | |
| height=380, | |
| ) | |
| analyse_btn = gr.Button("INITIALIZE SCAN & SYNTHESIZE AUDIO", elem_id="analyse-btn") | |
| # RIGHT COLUMN (Outputs - scrollable on desktop, fits on one screen) | |
| with gr.Column(scale=6, elem_classes="right-panel"): | |
| audio_out = gr.Audio( | |
| label="", | |
| type="filepath", | |
| autoplay=True, | |
| elem_id="audio-out", | |
| show_label=False | |
| ) | |
| concise_out = gr.HTML() | |
| detailed_out = gr.HTML() | |
| disclaimer_out = gr.HTML() | |
| # Link button logic | |
| analyse_btn.click( | |
| fn=analyse_report, | |
| inputs=[image_input], | |
| outputs=[detailed_out, concise_out, audio_out, disclaimer_out], | |
| ) | |
| if __name__ == "__main__": | |
| demo.launch(server_name="0.0.0.0", server_port=7860) |