Initial commit

.gitattributes

@@ -0,0 +1,5 @@
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.svg filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,16 @@
+# --- Backend (Flask) ---
+FROM python:3.9-slim
+WORKDIR /app
+
+# Copy backend files
+COPY app.py requirements.txt README.md ./
+COPY models/ ./models/
+
+# Copy static assets (brain examples, etc.) if present
+COPY static/ ./static/
+
+# Install Python deps
+RUN pip install --no-cache-dir -r requirements.txt
+
+ENV PORT=7860
+CMD ["gunicorn", "-b", "0.0.0.0:7860", "app:app"]

README.md

@@ -0,0 +1,8 @@
+---
+title: CerebroScan
+emoji: 🧠
+colorFrom: indigo
+colorTo: blue
+sdk: docker
+pinned: false
+---

app.py

@@ -0,0 +1,193 @@
+import os
+import re
+from dataclasses import dataclass
+from typing import Dict, List, Tuple
+
+import numpy as np
+from flask import Flask, jsonify, request
+from flask_cors import CORS
+
+import tensorflow as tf
+from tensorflow.keras.models import load_model
+
+
+# ----------------------------
+# Model definitions
+# ----------------------------
+
+@dataclass(frozen=True)
+class ModelSpec:
+    id: str
+    display_name: str          # what the user sees (friendly + technical)
+    filename: str              # under ./models/
+    arch: str                  # "resnet" | "efficientnet"
+    img_size: int              # input resolution
+    class_names: Tuple[str, ...]  # output order used during training
+    recommended_threshold: float  # per-model uncertainty cutoff (from notebooks)
+
+
+# NOTE:
+# Your training notebooks use *different* class ordering between the ResNet notebooks
+# (sorted unique categories) and the EfficientNet notebook (explicit list).
+# We keep per-model class order to avoid mislabeling probabilities.
+RESNET_CLASS_ORDER = ("MildDemented", "ModerateDemented", "NonDemented", "VeryMildDemented")
+EFFICIENTNET_CLASS_ORDER = ("NonDemented", "VeryMildDemented", "MildDemented", "ModerateDemented")
+
+MODEL_SPECS: List[ModelSpec] = [
+    ModelSpec("atlas", "Atlas — ResNet-50", "resnet50.h5", "resnet", 224, RESNET_CLASS_ORDER, 0.95),
+    ModelSpec("orion", "Orion — ResNet-101", "resnet101.h5", "resnet", 224, RESNET_CLASS_ORDER, 0.95),
+    ModelSpec("pulse", "Pulse — EfficientNet-B2", "efficientnetb2.h5", "efficientnet", 260, EFFICIENTNET_CLASS_ORDER, 0.95),
+]
+
+
+
+# ----------------------------
+# Flask app
+# ----------------------------
+
+app = Flask(__name__)
+CORS(app, resources={r"/api/*": {"origins": "*"}})
+
+# Lazy-loaded models (load on first use). Keep only what we need in CPU Spaces.
+_loaded_models: Dict[str, tf.keras.Model] = {}
+
+
+def _get_spec(model_id: str) -> ModelSpec:
+    for s in MODEL_SPECS:
+        if s.id == model_id:
+            return s
+    raise KeyError(f"Unknown model_id: {model_id}")
+
+
+def _get_preprocess_fn(arch: str):
+    if arch == "resnet":
+        from tensorflow.keras.applications.resnet50 import preprocess_input as resnet_preprocess
+        return resnet_preprocess
+    if arch == "efficientnet":
+        from tensorflow.keras.applications.efficientnet import preprocess_input as eff_preprocess
+        return eff_preprocess
+    raise ValueError(f"Unknown arch: {arch}")
+
+
+def _load_model(spec: ModelSpec) -> tf.keras.Model:
+    if spec.id in _loaded_models:
+        return _loaded_models[spec.id]
+
+    model_path = os.path.join(os.path.dirname(__file__), "models", spec.filename)
+    if not os.path.exists(model_path):
+        raise FileNotFoundError(
+            f"Model file not found: {model_path}. "
+            f"Place it at models/{spec.filename} in your Space."
+        )
+
+    # CPU-friendly TF settings (small wins on free Spaces)
+    try:
+        tf.config.threading.set_intra_op_parallelism_threads(0)
+        tf.config.threading.set_inter_op_parallelism_threads(0)
+    except Exception:
+        pass
+
+    model = load_model(model_path, compile=False)
+    _loaded_models[spec.id] = model
+    return model
+
+
+def _read_image(file_storage, img_size: int, preprocess_fn):
+    # Decode image
+    raw = file_storage.read()
+    image = tf.io.decode_image(raw, channels=3, expand_animations=False)
+    image = tf.image.resize(image, [img_size, img_size])
+    image = tf.cast(image, tf.float32)
+    image = preprocess_fn(image)
+    image = tf.expand_dims(image, axis=0)  # [1, H, W, 3]
+    return image
+
+
+def _predict(model: tf.keras.Model, image_tensor, class_names: Tuple[str, ...], threshold: float):
+    probs = model.predict(image_tensor, verbose=0)[0].astype(float)
+    probs = np.clip(probs, 0.0, 1.0)
+
+    best_idx = int(np.argmax(probs))
+    best_prob = float(np.max(probs))
+
+    # Add "Uncertain" post-hoc (not a model output class)
+    is_uncertain = best_prob < threshold
+
+    # Build response payload
+    by_class = [
+        {"id": name, "label": _pretty_label(name), "prob": float(probs[i])}
+        for i, name in enumerate(class_names)
+    ]
+    by_class.sort(key=lambda x: x["prob"], reverse=True)
+
+    return {
+        "prediction": {
+            "id": "Uncertain" if is_uncertain else class_names[best_idx],
+            "label": "Uncertain" if is_uncertain else _pretty_label(class_names[best_idx]),
+            "confidence": best_prob,
+            "threshold": threshold,
+        },
+        "probabilities": by_class,
+    }
+
+
+def _pretty_label(name: str) -> str:
+    # Internal training labels -> user-facing labels (final wording)
+    mapping = {
+        "NonDemented": "Healthy",
+        "VeryMildDemented": "Very Mildly Demented",
+        "MildDemented": "Mildly Demented",
+        "ModerateDemented": "Moderately Demented",
+        # Post-hoc
+        "Uncertain": "Uncertain",
+    }
+    return mapping.get(name, name)
+
+
+@app.get("/api/models")
+def api_models():
+    return jsonify({
+        "models": [
+            {
+                "id": s.id,
+                "name": s.display_name,
+                "img_size": s.img_size,
+                "classes": [{"id": c, "label": _pretty_label(c)} for c in s.class_names],
+                "recommended_threshold": s.recommended_threshold,
+            }
+            for s in MODEL_SPECS
+        ],
+        "default_model_id": MODEL_SPECS[0].id,
+    })
+
+
+@app.post("/api/classify")
+def api_classify():
+    if "file" not in request.files:
+        return jsonify({"error": "No file uploaded (field name must be 'file')."}), 400
+
+    model_id = request.form.get("model_id", MODEL_SPECS[0].id)
+    spec = _get_spec(model_id)
+    # Threshold is model-specific and not user-adjustable
+    threshold = spec.recommended_threshold
+
+    try:
+        model = _load_model(spec)
+        preprocess_fn = _get_preprocess_fn(spec.arch)
+
+        image_tensor = _read_image(request.files["file"], spec.img_size, preprocess_fn)
+        payload = _predict(model, image_tensor, spec.class_names, threshold)
+
+        payload["model"] = {"id": spec.id, "name": spec.display_name}
+        return jsonify(payload)
+
+    except FileNotFoundError as e:
+        return jsonify({"error": str(e)}), 500
+    except Exception as e:
+        return jsonify({"error": f"Failed to classify image: {e}"}), 500
+
+
+if __name__ == "__main__":
+    # Local dev: python app.py
+    # In Spaces (Dockerfile), gunicorn is used.
+    app.run(host="0.0.0.0", port=int(os.getenv("PORT", "7860")), debug=False)

models/README.md

@@ -0,0 +1,4 @@
+Place your model files here:
+- resnet50.h5
+- resnet101.h5
+- efficientnetb2.h5

models/efficientnetb2.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92ca0621bacfb11477e4242a4d38409b4f44a33483b5167dc5808e3413f7e243
+size 102821032

models/resnet101.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33946fb58dc03e58886e2b40fabff8ce545c6d75808b3ac87ab7d6b4cd75d39a
+size 524986352

models/resnet50.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b892a36a277c2210a545eb8743425188a3ab8893196512be55ac5d7250dec445
+size 296838384

requirements.txt

@@ -0,0 +1,7 @@
+flask
+flask-cors
+tensorflow-cpu
+numpy
+pillow
+h5py
+gunicorn

static/assets/brain/PLACE_IMAGES_HERE.txt

@@ -0,0 +1,3 @@
+Put these files here:
+- supportedexample.jpg
+- unsupportedexample.jpg

templates/index.html

@@ -0,0 +1,451 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8" />
+<title>CerebroScan</title>
+<meta name="viewport" content="width=device-width, initial-scale=1.0" />
+
+<style>
+:root{
+  --bg:#0b1220;
+  --card:#0f1b33;
+  --card2:#0c172e;
+  --text:#e8eefc;
+  --muted:#a9b7d0;
+  --border:rgba(255,255,255,.12);
+  --primary:#4f7cff;
+  --ring:rgba(79,124,255,.35);
+}
+
+:root[data-theme="light"]{
+  --bg:#f6f7fb;
+  --card:#ffffff;
+  --card2:#ffffff;
+  --text:#0c1222;
+  --muted:#44506a;
+  --border:rgba(0,0,0,.12);
+  --primary:#355dff;
+  --ring:rgba(53,93,255,.25);
+}
+
+*{box-sizing:border-box}
+body{
+  margin:0;
+  font-family:system-ui,-apple-system,Segoe UI,Roboto,sans-serif;
+  background:var(--bg);
+  color:var(--text);
+}
+
+header{
+  display:flex;
+  justify-content:space-between;
+  align-items:center;
+  padding:20px 28px;
+}
+
+.brand{display:flex;gap:12px;align-items:center}
+.brand img{width:40px;height:40px}
+
+.chip{
+  padding:8px 12px;
+  border-radius:999px;
+  background:var(--card);
+  border:1px solid var(--border);
+  cursor:pointer;
+}
+
+nav{
+  padding:0 28px 18px;
+  display:flex;
+  gap:10px;
+}
+
+.tab{
+  padding:8px 16px;
+  border-radius:999px;
+  border:1px solid var(--border);
+  background:rgba(255,255,255,.06);
+  cursor:pointer;
+}
+
+.tab.active{
+  background:rgba(79,124,255,.18);
+  box-shadow:0 0 0 4px var(--ring);
+}
+
+.section{display:none}
+.section.show{display:block}
+
+main{
+  display:grid;
+  grid-template-columns:1.2fr 1fr;
+  gap:24px;
+  padding:0 28px 28px;
+}
+@media(max-width:900px){main{grid-template-columns:1fr}}
+
+.card{
+  background:linear-gradient(180deg,var(--card),var(--card2));
+  border-radius:18px;
+  border:1px solid var(--border);
+  padding:22px;
+}
+
+h2{margin:0 0 12px;font-size:20px}
+
+.controls{
+  display:flex;
+  gap:12px;
+  flex-wrap:wrap;
+  align-items:center;
+}
+
+.select-wrap{position:relative}
+.select-wrap::after{
+  content:"▾";
+  position:absolute;
+  right:14px;
+  top:50%;
+  transform:translateY(-50%);
+  pointer-events:none;
+  color:var(--muted);
+}
+
+select{
+  appearance:none;
+  padding:10px 40px 10px 14px;
+  border-radius:12px;
+  border:1px solid var(--border);
+  background:rgba(255,255,255,.06);
+  color:var(--text);
+  font-weight:600;
+  min-width:260px;
+}
+
+.file-btn{
+  padding:10px 14px;
+  border-radius:12px;
+  border:1px dashed var(--border);
+  cursor:pointer;
+  background:rgba(255,255,255,.04);
+}
+
+button.primary{
+  padding:10px 18px;
+  border-radius:12px;
+  border:none;
+  background:linear-gradient(180deg,var(--primary),#2f62ff);
+  color:white;
+  font-weight:700;
+  cursor:pointer;
+}
+
+button.secondary{
+  padding:10px 14px;
+  border-radius:12px;
+  border:1px solid var(--border);
+  background:rgba(255,255,255,.04);
+  cursor:pointer;
+}
+
+button:disabled{opacity:.6}
+
+.upload-box{
+  margin-top:14px;
+  padding:18px;
+  border:2px dashed var(--border);
+  border-radius:14px;
+  text-align:center;
+  color:var(--muted);
+}
+
+.preview{margin-top:14px;background:black;border-radius:14px;overflow:hidden}
+.preview.hidden{display:none}
+.preview img{width:100%;max-height:360px;object-fit:contain}
+
+.result{margin-top:16px}
+.result-title{font-size:22px;font-weight:800}
+.result-msg{color:var(--muted);margin-top:6px}
+
+.examples .ex{
+  margin-top:12px;
+  border:1px solid var(--border);
+  border-radius:14px;
+  overflow:hidden;
+}
+.examples .lbl{padding:10px;font-weight:800;border-bottom:1px solid var(--border)}
+.examples .desc{padding:0 10px 10px;color:var(--muted)}
+.examples img{width:100%;max-height:280px;object-fit:contain;background:black}
+
+.team{
+  display:grid;
+  grid-template-columns:repeat(auto-fit,minmax(240px,1fr));
+  gap:14px;
+}
+.member{
+  display:flex;
+  gap:12px;
+  align-items:center;
+  padding:10px;
+  border:1px solid var(--border);
+  border-radius:14px;
+}
+.member img{width:48px;height:48px;border-radius:50%;cursor:pointer}
+
+footer{
+  padding:14px 28px;
+  border-top:1px solid var(--border);
+  font-size:13px;
+  color:var(--muted);
+}
+
+/* modal */
+.modal-backdrop{
+  position:fixed;
+  inset:0;
+  background:rgba(0,0,0,.6);
+  display:none;
+  align-items:center;
+  justify-content:center;
+}
+.modal-backdrop.show{display:flex}
+.modal{
+  background:var(--card);
+  padding:18px;
+  border-radius:16px;
+  border:1px solid var(--border);
+  width:420px;
+}
+</style>
+</head>
+
+<body>
+
+<header>
+  <div class="brand">
+    <img src="{{ url_for('static', filename='assets/brain/brain.svg') }}">
+    <div>
+      <strong>CerebroScan</strong><br>
+      <small style="color:var(--muted)">AI-assisted MRI screening (demo)</small>
+    </div>
+  </div>
+  <button id="themeToggle" class="chip">☀️</button>
+</header>
+
+<nav>
+  <button class="tab active" onclick="showSection('scan',this)">Scan</button>
+  <button class="tab" onclick="showSection('info',this)">Info</button>
+</nav>
+
+<section id="scan" class="section show">
+<main>
+<div class="card">
+<h2>New scan</h2>
+
+<div class="controls">
+  <span class="select-wrap"><select id="modelSelect"></select></span>
+
+  <label class="file-btn">
+    Choose image
+    <input type="file" id="fileInput" accept="image/*" hidden>
+  </label>
+
+  <button class="primary" id="runBtn" onclick="runScan()">Run scan</button>
+  <button class="secondary" id="newBtn" style="display:none" onclick="newScan()">New scan</button>
+</div>
+
+<div class="upload-box" id="dropZone">
+Drag & drop or paste (Ctrl+V) an MRI image
+</div>
+
+<div class="preview hidden" id="previewBox">
+<img id="previewImg">
+</div>
+
+<div class="result">
+<div id="resultTitle" class="result-title">—</div>
+<div id="resultMsg" class="result-msg">Upload an image to begin.</div>
+<button id="likelyBtn" class="secondary" style="display:none;margin-top:10px" onclick="showLikely()">Show most likely result</button>
+</div>
+</div>
+
+<div class="card">
+<h2>Examples</h2>
+<div class="examples">
+  <div class="ex">
+    <div class="lbl">Supported example</div>
+    <div class="desc">Brain-only MRI slice (no skull visible).</div>
+    <img src="{{ url_for('static', filename='assets/brain/supportedexample.jpg') }}">
+  </div>
+  <div class="ex">
+    <div class="lbl">Unsupported example</div>
+    <div class="desc">Skull visible — please upload a brain-only slice.</div>
+    <img src="{{ url_for('static', filename='assets/brain/unsupportedexample.jpg') }}">
+  </div>
+</div>
+</div>
+</main>
+</section>
+
+<section id="info" class="section">
+<main>
+<div class="card">
+<h2>About CerebroScan</h2>
+<p>Educational demo for Alzheimer’s stage classification. Low confidence → <b>Uncertain</b>.</p>
+
+<h2>Under the supervision of:</h2>
+<p>
+Prof. Muhammad Sayed Hammad<br>
+Eng. Heidi Ahmed
+</p>
+
+<h2>Team</h2>
+<div class="team">
+<div class="member"><img src="{{ url_for('static', filename='assets/images/team/fatma.jpg') }}" onclick="window.open('https://www.linkedin.com/in/fatma-al-zahraa-emad-326b64234/')"><span>Fatma Al-Zahraa Emad</span></div>
+<div class="member"><img src="{{ url_for('static', filename='assets/images/team/gehad.jpg') }}" onclick="window.open('https://www.linkedin.com/in/gehad-mohamed-2a4946252/')"><span>Gehad Mohamed</span></div>
+<div class="member"><img src="{{ url_for('static', filename='assets/images/team/heba.jpg') }}" onclick="window.open('https://www.linkedin.com/in/hebatullah-elgazoly-308ab2243/')"><span>Hebatullah El Gazoly</span></div>
+<div class="member"><img src="{{ url_for('static', filename='assets/images/team/asem.jpg') }}" onclick="window.open('https://www.linkedin.com/in/mohamedasem318/')"><span>Mohamed Assem</span></div>
+<div class="member"><img src="{{ url_for('static', filename='assets/images/team/sameh.jpg') }}" onclick="window.open('https://www.linkedin.com/in/muhamedsameh/')"><span>Mohamed Sameh</span></div>
+</div>
+</div>
+</main>
+</section>
+
+<footer>
+Contact: <a href="mailto:mohamedasem318@gmail.com">Mohamed Assem</a> •
+<a href="mailto:mohamed.sameh8103@gmail.com">Mohamed Sameh</a><br>
+Educational demo — not medical advice
+</footer>
+
+<div class="modal-backdrop" id="modalBackdrop" onclick="closeModalIfBackdrop(event)">
+<div class="modal">
+<h3>Most likely result</h3>
+<p id="modalLabel"></p>
+<p id="modalProb"></p>
+<button class="secondary" onclick="closeModal()">Close</button>
+</div>
+</div>
+
+<script>
+function showSection(id,btn){
+document.querySelectorAll('.section').forEach(s=>s.classList.remove('show'));
+document.getElementById(id).classList.add('show');
+document.querySelectorAll('.tab').forEach(t=>t.classList.remove('active'));
+btn.classList.add('active');
+}
+
+function setTheme(t){
+document.documentElement.dataset.theme=t;
+localStorage.setItem("theme",t);
+themeToggle.textContent=t==="dark"?"☀️":"🌙";
+}
+themeToggle.onclick=()=>setTheme(document.documentElement.dataset.theme==="dark"?"light":"dark");
+setTheme(localStorage.getItem("theme")||"dark");
+
+let currentFile=null,lastMostLikely=null;
+const previewBox=document.getElementById("previewBox");
+const previewImg=document.getElementById("previewImg");
+
+function setFile(f){
+currentFile=f;
+previewImg.src=URL.createObjectURL(f);
+previewBox.classList.remove("hidden");
+resultTitle.textContent="—";
+resultMsg.textContent="Ready to run scan.";
+likelyBtn.style.display="none";
+newBtn.style.display="none";
+lastMostLikely=null;
+}
+
+fileInput.onchange=e=>e.target.files[0]&&setFile(e.target.files[0]);
+
+dropZone.ondragover=e=>{e.preventDefault()};
+dropZone.ondrop=e=>{
+e.preventDefault();
+e.dataTransfer.files[0]&&setFile(e.dataTransfer.files[0]);
+};
+
+window.addEventListener("paste",e=>{
+for(const i of e.clipboardData.items){
+if(i.type.startsWith("image/")){setFile(i.getAsFile());break;}
+}
+});
+
+async function loadModels(){
+const r=await fetch("/api/models");
+const d=await r.json();
+modelSelect.innerHTML="";
+d.models.forEach(m=>{
+const o=document.createElement("option");
+o.value=m.id;o.textContent=m.name;
+modelSelect.appendChild(o);
+});
+modelSelect.value=d.default_model_id;
+}
+
+async function runScan(){
+if(!currentFile){resultMsg.textContent="Upload an image first.";return;}
+runBtn.disabled=true;
+modelSelect.disabled=true;
+resultTitle.textContent="Running…";
+resultMsg.textContent="Analyzing image…";
+likelyBtn.style.display="none";
+lastMostLikely=null;
+
+try{
+const fd=new FormData();
+fd.append("file",currentFile);
+fd.append("model_id",modelSelect.value);
+const r=await fetch("/api/classify",{method:"POST",body:fd});
+const d=await r.json();
+
+if(d.prediction.label==="Uncertain"){
+resultTitle.textContent="Uncertain";
+resultMsg.textContent="Consult a professional.";
+const probs=[...(d.probabilities||[])].sort((a,b)=>b.prob-a.prob);
+if(probs[0]){
+lastMostLikely=probs[0];
+likelyBtn.style.display="inline-block";
+}
+}else{
+resultTitle.textContent=d.prediction.label;
+resultMsg.textContent=`Confidence: ${(d.prediction.confidence*100).toFixed(1)}%`;
+likelyBtn.style.display="none";
+lastMostLikely=null;
+}
+newBtn.style.display="inline-block";
+}catch(e){
+resultTitle.textContent="Sorry";
+resultMsg.textContent="Something went wrong.";
+}finally{
+runBtn.disabled=false;
+modelSelect.disabled=false;
+}
+}
+
+function newScan(){
+currentFile=null;
+fileInput.value="";
+previewBox.classList.add("hidden");
+previewImg.src="";
+resultTitle.textContent="—";
+resultMsg.textContent="Upload an image to begin.";
+likelyBtn.style.display="none";
+newBtn.style.display="none";
+lastMostLikely=null;
+}
+
+function showLikely(){
+if(!lastMostLikely)return;
+modalLabel.textContent=lastMostLikely.label;
+modalProb.textContent=`Probability: ${(lastMostLikely.prob*100).toFixed(1)}%`;
+modalBackdrop.classList.add("show");
+}
+function closeModal(){modalBackdrop.classList.remove("show")}
+function closeModalIfBackdrop(e){if(e.target.id==="modalBackdrop")closeModal()}
+
+loadModels();
+</script>
+
+</body>
+</html>