app.py

import streamlit as st
import numpy as np
import cv2
import os
import random
import tempfile
import joblib
from typing import Optional, Tuple, Dict, Any, List

# NEW: use components.html for the long HTML report
import streamlit.components.v1 as components

# (Optional) Face landmarks similar to earlier app
try:
    import mediapipe as mp
    mp_face_mesh = mp.solutions.face_mesh
    face_mesh = mp_face_mesh.FaceMesh(static_image_mode=True, max_num_faces=1,
                                      refine_landmarks=True, min_detection_confidence=0.5)
except Exception:
    mp = None
    face_mesh = None

# Live video capture (front-end) with WebRTC
from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration
import av
from collections import deque

st.set_page_config(page_title="Face-based Lab Test AI Report", layout="wide")

# ==============================
# WebRTC config
# ==============================
RTC_CONFIGURATION = RTCConfiguration(
    {"iceServers": [{"urls": ["stun:stun.l.google.com:19302"]}]}
)

# ==============================
# Utility & UI helpers (copied style)
# ==============================
def get_risk_color(value: float, normal_range: Tuple[float, float]) -> Tuple[str, str, str]:
    low, high = normal_range
    if value < low:
        return ("Low", "🔻", "#FFCCCC")
    elif value > high:
        return ("High", "🔺", "#FFE680")
    else:
        return ("Normal", "✅", "#CCFFCC")

def build_table(title: str, rows: List[Tuple[str, float, Tuple[float, float]]]) -> str:
    html = (
        f'<div style="margin-bottom: 24px;">'
        f'<h4 style="margin: 8px 0;">{title}</h4>'
        f'<table style="width:100%; border-collapse:collapse;">'
        f'<thead><tr style="background:#f0f0f0;">'
        f'<th style="padding:8px;border:1px solid #ccc;">Test</th>'
        f'<th style="padding:8px;border:1px solid #ccc;">Result</th>'
        f'<th style="padding:8px;border:1px solid #ccc;">Expected Range</th>'
        f'<th style="padding:8px;border:1px solid #ccc;">Level</th>'
        f'</tr></thead><tbody>'
    )
    for label, value, ref in rows:
        level, icon, bg = get_risk_color(value, ref)
        html += (
            f'<tr style="background:{bg};">'
            f'<td style="padding:6px;border:1px solid #ccc;">{label}</td>'
            f'<td style="padding:6px;border:1px solid #ccc;">{value:.2f}</td>'
            f'<td style="padding:6px;border:1px solid #ccc;">{ref[0]} – {ref[1]}</td>'
            f'<td style="padding:6px;border:1px solid #ccc;">{icon} {level}</td>'
            f'</tr>'
        )
    html += '</tbody></table></div>'
    return html

def extract_basic_rgb_features(image_rgb: np.ndarray) -> List[float]:
    red_percent = 100 * float(np.mean(image_rgb[:, :, 0])) / 255.0
    green_percent = 100 * float(np.mean(image_rgb[:, :, 1])) / 255.0
    blue_percent = 100 * float(np.mean(image_rgb[:, :, 2])) / 255.0
    return [red_percent, green_percent, blue_percent]

def blur_for_display(img_rgb: np.ndarray, k: int = 35) -> np.ndarray:
    k = max(3, k | 1)  # odd kernel
    return cv2.GaussianBlur(img_rgb, (k, k), 0)

# ==============================
# Model loading (guarded)
# ==============================
def safe_load_joblib(path: str):
    try:
        if os.path.exists(path):
            return joblib.load(path)
    except Exception:
        pass
    return None

hemoglobin_model = safe_load_joblib("hemoglobin_model_from_anemia_dataset.pkl")
hr_model = safe_load_joblib("heart_rate_model.pkl")
spo2_model = safe_load_joblib("spo2_model_simulated.pkl")
hemoglobin_r2 = 0.385  # as shown in your previous UI

# ==============================
# Placeholder models
# ==============================
def random_model_predict(low: float, high: float) -> float:
    return random.uniform(low, high)

# ==============================
# Image path (analyze image once, show only final blurred frame)
# ==============================
def analyze_face(image_bgr: Optional[np.ndarray]) -> Tuple[str, Optional[np.ndarray]]:
    if image_bgr is None:
        return "<div style='color:red;'>⚠️ Error: No image provided.</div>", None

    frame_rgb = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)

    if face_mesh is not None and mp is not None:
        result = face_mesh.process(frame_rgb)
        if not result.multi_face_landmarks:
            return "<div style='color:red;'>⚠️ Error: Face not detected.</div>", None

    features = extract_basic_rgb_features(frame_rgb)

    if hemoglobin_model is not None:
        try:
            hb_pred = float(hemoglobin_model.predict([features])[0])
        except Exception:
            hb_pred = float(random_model_predict(12.5, 15.0))
    else:
        hb_pred = float(random_model_predict(12.5, 15.0))

    heart_rate = float(random_model_predict(60, 100))
    spo2 = float(random_model_predict(95, 100))
    rr = int(12 + abs(int(heart_rate) % 5 - 2))

    test_values: Dict[str, Any] = {
        "Hemoglobin": hb_pred,
        "WBC Count": random_model_predict(4.0, 11.0),
        "Platelet Count": random_model_predict(150, 450),
        "Iron": random_model_predict(60, 170),
        "Ferritin": random_model_predict(30, 300),
        "TIBC": random_model_predict(250, 400),
        "Bilirubin": random_model_predict(0.3, 1.2),
        "Creatinine": random_model_predict(0.6, 1.2),
        "Urea": random_model_predict(7, 20),
        "Sodium": random_model_predict(135, 145),
        "Potassium": random_model_predict(3.5, 5.1),
        "TSH": random_model_predict(0.4, 4.0),
        "Cortisol": random_model_predict(5, 25),
        "FBS": random_model_predict(70, 110),
        "HbA1c": random_model_predict(4.0, 5.7),
        "Albumin": random_model_predict(3.5, 5.5),
        "BP Systolic": random_model_predict(90, 120),
        "BP Diastolic": random_model_predict(60, 80),
        "Temperature": random_model_predict(97, 99),
    }

    html_output = "".join([
        f'<div style="font-size:14px;color:#888;margin-bottom:10px;">Hemoglobin R² Score: {hemoglobin_r2:.2f}</div>',
        build_table("🩸 Hematology", [
            ("Hemoglobin", test_values["Hemoglobin"], (13.5, 17.5)),
            ("WBC Count", test_values["WBC Count"], (4.0, 11.0)),
            ("Platelet Count", test_values["Platelet Count"], (150, 450)),
        ]),
        build_table("🧬 Iron Panel", [
            ("Iron", test_values["Iron"], (60, 170)),
            ("Ferritin", test_values["Ferritin"], (30, 300)),
            ("TIBC", test_values["TIBC"], (250, 400)),
        ]),
        build_table("🧬 Liver & Kidney", [
            ("Bilirubin", test_values["Bilirubin"], (0.3, 1.2)),
            ("Creatinine", test_values["Creatinine"], (0.6, 1.2)),
            ("Urea", test_values["Urea"], (7, 20)),
        ]),
        build_table("🧪 Electrolytes", [
            ("Sodium", test_values["Sodium"], (135, 145)),
            ("Potassium", test_values["Potassium"], (3.5, 5.1)),
        ]),
        build_table("🧁 Metabolic & Thyroid", [
            ("FBS", test_values["FBS"], (70, 110)),
            ("HbA1c", test_values["HbA1c"], (4.0, 5.7)),
            ("TSH", test_values["TSH"], (0.4, 4.0)),
        ]),
        build_table("❤️ Vitals", [
            ("SpO2", spo2, (95, 100)),
            ("Heart Rate", heart_rate, (60, 100)),
            ("Respiratory Rate", rr, (12, 20)),
            ("Temperature", test_values["Temperature"], (97, 99)),
            ("BP Systolic", test_values["BP Systolic"], (90, 120)),
            ("BP Diastolic", test_values["BP Diastolic"], (60, 80)),
        ]),
        build_table("🩹 Other Indicators", [
            ("Cortisol", test_values["Cortisol"], (5, 25)),
            ("Albumin", test_values["Albumin"], (3.5, 5.5)),
        ])
    ])

    summary = "<div style='margin-top:20px;padding:12px;border:1px dashed #999;background:#fcfcfc;'>"
    summary += "<h4>📝 Summary for You</h4><ul>"
    if hb_pred < 13.5:
        summary += "<li>Your hemoglobin is a bit low — this could mean mild anemia.</li>"
    if test_values["Iron"] < 60 or test_values["Ferritin"] < 30:
        summary += "<li>Low iron storage detected — consider an iron profile test.</li>"
    if test_values["Bilirubin"] > 1.2:
        summary += "<li>Elevated bilirubin — possible jaundice. Recommend LFT.</li>"
    if test_values["HbA1c"] > 5.7:
        summary += "<li>High HbA1c — prediabetes indication. Recommend glucose check.</li>"
    if spo2 < 95:
        summary += "<li>Low SpO₂ — suggest retesting with a pulse oximeter.</li>"
    summary += "</ul><p><strong>💡 Tip:</strong> This is an AI-based estimate. Please follow up with a lab.</p></div>"
    html_output += summary

    html_output += "<br><div style='margin-top:20px;padding:12px;border:2px solid #2d87f0;background:#f2faff;text-align:center;border-radius:8px;'>"
    html_output += "<h4>📞 Book a Lab Test</h4><p>Prefer confirmation? Find certified labs near you.</p>"
    html_output += "<button style='padding:10px 20px;background:#007BFF;color:#fff;border:none;border-radius:5px;cursor:pointer;'>Find Labs Near Me</button></div>"

    # Return the single final frame (blurred for display only)
    return html_output, blur_for_display(frame_rgb)

# ==============================
# Video path (assemble once; show only final blurred keyframe)
# ==============================
def analyze_video(video_path: Optional[str]) -> Tuple[str, Optional[np.ndarray]]:
    if not video_path or not os.path.exists(video_path):
        return "<div style='color:red;'>⚠️ Face video missing or unreadable.</div>", None

    cap = cv2.VideoCapture(video_path)
    frame_sample = None
    frames = 0
    while True:
        ret, frame = cap.read()
        if not ret:
            break
        if frame_sample is None:
            frame_sample = frame.copy()
        frames += 1
    cap.release()

    if frame_sample is None or frames < 30:
        return "<div style='color:red;'>⚠️ Video too short; please record ~20–30s.</div>", None

    # Placeholder HR/SpO2 (replace with proper rPPG)
    try:
        hr_value = float(random_model_predict(65, 85))
        gray = cv2.cvtColor(frame_sample, cv2.COLOR_BGR2GRAY)
        brightness_var = float(np.std(gray) / 255.0)
        skin_tone_index = float(np.mean(frame_sample[100:150, 100:150]) / 255.0) if frame_sample[100:150, 100:150].size else 0.5
        spo2 = float(98.0 - (1.5 * brightness_var - 0.5 * (skin_tone_index - 0.5)))
        spo2 = float(np.clip(spo2, 92.0, 100.0))
    except Exception:
        hr_value = float(random_model_predict(65, 85))
        spo2 = float(random_model_predict(95, 100))

    rr = int(12 + abs(int(hr_value) % 5 - 2))

    test_values: Dict[str, Any] = {
        "Hemoglobin": random_model_predict(12.5, 15.0),
        "WBC Count": random_model_predict(4.0, 11.0),
        "Platelet Count": random_model_predict(150, 450),
        "Iron": random_model_predict(60, 170),
        "Ferritin": random_model_predict(30, 300),
        "TIBC": random_model_predict(250, 400),
        "Bilirubin": random_model_predict(0.3, 1.2),
        "Creatinine": random_model_predict(0.6, 1.2),
        "Urea": random_model_predict(7, 20),
        "Sodium": random_model_predict(135, 145),
        "Potassium": random_model_predict(3.5, 5.1),
        "TSH": random_model_predict(0.4, 4.0),
        "Cortisol": random_model_predict(5, 25),
        "FBS": random_model_predict(70, 110),
        "HbA1c": random_model_predict(4.0, 5.7),
        "Albumin": random_model_predict(3.5, 5.5),
        "BP Systolic": random_model_predict(90, 120),
        "BP Diastolic": random_model_predict(60, 80),
        "Temperature": random_model_predict(97, 99),
    }

    frame_rgb = cv2.cvtColor(frame_sample, cv2.COLOR_BGR2RGB)

    html_output = "".join([
        f'<div style="font-size:14px;color:#888;margin-bottom:10px;">Hemoglobin R² Score: {hemoglobin_r2:.2f}</div>',
        build_table("🩸 Hematology", [
            ("Hemoglobin", test_values["Hemoglobin"], (13.5, 17.5)),
            ("WBC Count", test_values["WBC Count"], (4.0, 11.0)),
            ("Platelet Count", test_values["Platelet Count"], (150, 450)),
        ]),
        build_table("🧬 Iron Panel", [
            ("Iron", test_values["Iron"], (60, 170)),
            ("Ferritin", test_values["Ferritin"], (30, 300)),
            ("TIBC", test_values["TIBC"], (250, 400)),
        ]),
        build_table("🧬 Liver & Kidney", [
            ("Bilirubin", test_values["Bilirubin"], (0.3, 1.2)),
            ("Creatinine", test_values["Creatinine"], (0.6, 1.2)),
            ("Urea", test_values["Urea"], (7, 20)),
        ]),
        build_table("🧪 Electrolytes", [
            ("Sodium", test_values["Sodium"], (135, 145)),
            ("Potassium", test_values["Potassium"], (3.5, 5.1)),
        ]),
        build_table("🧁 Metabolic & Thyroid", [
            ("FBS", test_values["FBS"], (70, 110)),
            ("HbA1c", test_values["HbA1c"], (4.0, 5.7)),
            ("TSH", test_values["TSH"], (0.4, 4.0)),
        ]),
        build_table("❤️ Vitals", [
            ("SpO2", spo2, (95, 100)),
            ("Heart Rate", hr_value, (60, 100)),
            ("Respiratory Rate", rr, (12, 20)),
            ("Temperature", test_values["Temperature"], (97, 99)),
            ("BP Systolic", test_values["BP Systolic"], (90, 120)),
            ("BP Diastolic", test_values["BP Diastolic"], (60, 80)),
        ]),
        build_table("🩹 Other Indicators", [
            ("Cortisol", test_values["Cortisol"], (5, 25)),
            ("Albumin", test_values["Albumin"], (3.5, 5.5)),
        ])
    ])

    summary = "<div style='margin-top:20px;padding:12px;border:1px dashed #999;background:#fcfcfc;'>"
    summary += "<h4>📝 Summary for You</h4><ul>"
    if test_values["Hemoglobin"] < 13.5:
        summary += "<li>Your hemoglobin is a bit low — this could mean mild anemia.</li>"
    if test_values["Iron"] < 60 or test_values["Ferritin"] < 30:
        summary += "<li>Low iron storage detected — consider an iron profile test.</li>"
    if test_values["Bilirubin"] > 1.2:
        summary += "<li>Elevated bilirubin — possible jaundice. Recommend LFT.</li>"
    if test_values["HbA1c"] > 5.7:
        summary += "<li>High HbA1c — prediabetes indication. Recommend glucose check.</li>"
    if spo2 < 95:
        summary += "<li>Low SpO₂ — suggest retesting with a pulse oximeter.</li>"
    summary += "</ul><p><strong>💡 Tip:</strong> This is an AI-based estimate. Please follow up with a lab.</p></div>"
    html_output += summary

    html_output += "<br><div style='margin-top:20px;padding:12px;border:2px solid #2d87f0;background:#f2faff;text-align:center;border-radius:8px;'>"
    html_output += "<h4>📞 Book a Lab Test</h4><p>Prefer confirmation? Find certified labs near you.</p>"
    html_output += "<button style='padding:10px 20px;background:#007BFF;color:#fff;border:none;border-radius:5px;cursor:pointer;'>Find Labs Near Me</button></div>"

    return html_output, blur_for_display(frame_rgb)

# ==============================
# WebRTC frame collector (no progress timer to avoid constant re-renders)
# ==============================
class HRCollectorVideoProcessor:
    def __init__(self):
        self.frames = deque(maxlen=30 * 90)  # up to ~90s @30fps
        self.recording = False

    def recv(self, frame: av.VideoFrame) -> av.VideoFrame:
        img = frame.to_ndarray(format="bgr24")
        if self.recording:
            self.frames.append(img)
        return av.VideoFrame.from_ndarray(img, format="bgr24")

    def start(self):
        self.recording = True
        self.frames.clear()

    def stop_and_dump_to_file(self) -> Optional[str]:
        self.recording = False
        if len(self.frames) < 30:  # ~1s
            return None
        h, w = self.frames[0].shape[:2]
        fourcc = cv2.VideoWriter_fourcc(*"mp4v")
        tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
        tmp_path = tmp.name
        tmp.close()
        out = cv2.VideoWriter(tmp_path, fourcc, 30.0, (w, h))
        for f in self.frames:
            out.write(f)
        out.release()
        return tmp_path

# ==============================
# Streamlit UI
# ==============================
st.markdown("""
# 🧠 Face-Based Lab Test AI Report
Use **Image** (Hb) or **Video** (HR). Only a single **blurred preview** is shown **after** analysis to avoid flicker.
""")

# Session state to hold captured assets between form submits
if "captured_image_bgr" not in st.session_state:
    st.session_state.captured_image_bgr = None
if "captured_video_path" not in st.session_state:
    st.session_state.captured_video_path = None
# NEW: buffer raw camera bytes in case user forgets to click "Save Image"
if "img_bytes_buffer" not in st.session_state:
    st.session_state.img_bytes_buffer = None

mode = st.radio("Choose Input Mode", ["Image", "Video"], horizontal=True)

col_left, col_right = st.columns([1.2, 1], gap="large")

with col_left:
    if mode == "Image":
        st.subheader("📸 Face / Eye Image")

        # Wrap inputs in a form so widget changes don't re-run the app
        with st.form("img_form", clear_on_submit=False):
            img_source = st.radio("Source", ["Camera", "Upload"], horizontal=True, key="img_source")
            uploaded_img = None
            if img_source == "Camera":
                # KEY ADDED + auto-buffer captured bytes
                cam_img = st.camera_input("Capture image", key="camera_img")
                if cam_img is not None:
                    st.session_state.img_bytes_buffer = cam_img.getvalue()
            else:
                file_up = st.file_uploader("Upload JPG/PNG", type=["jpg", "jpeg", "png"])
                if file_up is not None:
                    uploaded_img = file_up.read()

            submitted_img = st.form_submit_button("Save Image")
            if submitted_img:
                # Prefer explicit upload; else fall back to auto-buffered camera bytes
                raw_bytes = uploaded_img if uploaded_img else st.session_state.img_bytes_buffer
                if raw_bytes:
                    arr = np.frombuffer(raw_bytes, dtype=np.uint8)
                    img_bgr = cv2.imdecode(arr, cv2.IMREAD_COLOR)
                    if img_bgr is not None:
                        st.session_state.captured_image_bgr = img_bgr
                        st.success("Image saved. Now click **Analyze Image** below.")
                    else:
                        st.warning("Could not decode the image. Please recapture/upload.")
                else:
                    st.warning("No image provided.")

        analyze_image = st.button("🔍 Analyze Image", type="primary", use_container_width=True)

    else:
        st.subheader("📽 Face Video")

        with st.form("vid_form", clear_on_submit=False):
            vid_source = st.radio("Source", ["Camera (Live)", "Upload"], horizontal=True, key="vid_source")
            temp_video_path = None

            if vid_source == "Upload":
                up_vid = st.file_uploader("Upload MP4/AVI/MOV", type=["mp4", "avi", "mov"])
                if up_vid is not None:
                    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as tmpf:
                        tmpf.write(up_vid.read())
                        temp_video_path = tmpf.name
            else:
                st.write("Start → record ~20–30s → Stop & Use")
                if "webrtc_ctx" not in st.session_state:
                    st.session_state.webrtc_ctx = None
                    st.session_state.hr_processor = HRCollectorVideoProcessor()

                ctx = webrtc_streamer(
                    key="hr-webrtc",
                    mode=WebRtcMode.SENDRECV,
                    rtc_configuration=RTC_CONFIGURATION,
                    media_stream_constraints={"video": True, "audio": False},
                    video_processor_factory=lambda: st.session_state.hr_processor,
                )

                c1, c2 = st.columns(2)
                with c1:
                    start_clicked = st.form_submit_button("Start Recording")
                with c2:
                    stop_clicked = st.form_submit_button("Stop & Use")

                if start_clicked and ctx.state.playing:
                    st.session_state.hr_processor.start()
                    st.info("Recording started...")

                if stop_clicked:
                    dump_path = st.session_state.hr_processor.stop_and_dump_to_file()
                    if dump_path and os.path.exists(dump_path):
                        temp_video_path = dump_path
                        st.success("Video captured. Now click **Analyze Video** below.")
                    else:
                        st.warning("Captured video too short. Please record ~20–30 seconds.")

            # Store selected/captured video after form submit
            submitted_vid = st.form_submit_button("Save Video")
            if submitted_vid:
                if temp_video_path:
                    st.session_state.captured_video_path = temp_video_path
                elif vid_source == "Upload":
                    st.warning("No video uploaded.")
                else:
                    st.warning("No video captured yet.")

        analyze_video_btn = st.button("🔍 Analyze Video", type="primary", use_container_width=True)

with col_right:
    st.subheader("🔒 Blurred Preview (shown only after analysis)")
    preview_placeholder = st.empty()

st.markdown("---")

# Single HTML report placeholder
report_placeholder = st.empty()

# Run analysis only when the explicit Analyze buttons are pressed
if mode == "Image" and 'analyze_image' in locals() and analyze_image:
    # If user forgot "Save Image", try auto-buffer
    if st.session_state.captured_image_bgr is None and st.session_state.img_bytes_buffer is not None:
        arr = np.frombuffer(st.session_state.img_bytes_buffer, dtype=np.uint8)
        img_bgr = cv2.imdecode(arr, cv2.IMREAD_COLOR)
        if img_bgr is not None:
            st.session_state.captured_image_bgr = img_bgr

    html, frame_rgb_blurred = analyze_face(st.session_state.captured_image_bgr)
    if frame_rgb_blurred is not None:
        preview_placeholder.image(frame_rgb_blurred, caption="Blurred Image", use_container_width=True)
    with report_placeholder:
        components.html(html, height=1200, scrolling=True)

elif mode == "Video" and 'analyze_video_btn' in locals() and analyze_video_btn:
    html, frame_rgb_blurred = analyze_video(st.session_state.captured_video_path)
    if frame_rgb_blurred is not None:
        preview_placeholder.image(frame_rgb_blurred, caption="Blurred Key Frame", use_container_width=True)
    with report_placeholder:
        components.html(html, height=1200, scrolling=True)