Upload app.py with huggingface_hub

app.py

@@ -17,6 +17,11 @@ _STEP         = 3
 _MAX_WINDOWS  = 30
 _FC_THRESHOLD = 0.2
 
+# CC200 atlas (Craddock 2012) → approximate Yeo 7-network parcellation
+_NET_NAMES  = ["DMN", "Salience", "Frontoparietal", "Sensorimotor", "Visual", "Dorsal Attn", "Subcortical"]
+_NET_BOUNDS = [0, 38, 69, 99, 137, 165, 180, 200]
+_NET_COLORS = ["#e63946", "#f4a261", "#457b9d", "#2dc653", "#a8dadc", "#8b5cf6", "#6b7280"]
+
 _CKPTS = {
     "NYU":  Path("checkpoints/nyu.ckpt"),
     "USM":  Path("checkpoints/usm.ckpt"),
@@ -89,44 +94,86 @@ def _saliency_figure(sal, p_mean):
     import matplotlib.pyplot as plt
     from PIL import Image
 
-    thresh  = np.percentile(sal, 95)
-    sal_top = np.where(sal >= thresh, sal, 0.0)
-    roi_imp = sal.sum(1)
-    top20   = roi_imp.argsort()[-20:][::-1]
-    color   = "#e63946" if p_mean > 0.6 else "#2dc653" if p_mean < 0.4 else "#f4a261"
+    n_nets = len(_NET_NAMES)
+
+    # Aggregate 200×200 saliency → 7×7 network-level matrix
+    net_sal = np.zeros((n_nets, n_nets))
+    for i, (s1, e1) in enumerate(zip(_NET_BOUNDS[:-1], _NET_BOUNDS[1:])):
+        for j, (s2, e2) in enumerate(zip(_NET_BOUNDS[:-1], _NET_BOUNDS[1:])):
+            net_sal[i, j] = sal[s1:e1, s2:e2].mean()
+
+    # Network importance: mean outgoing + incoming saliency per network
+    net_imp = np.array([
+        sal[s:e, :].mean() + sal[:, s:e].mean()
+        for s, e in zip(_NET_BOUNDS[:-1], _NET_BOUNDS[1:])
+    ])
 
-    fig, axes = plt.subplots(1, 2, figsize=(14, 5))
+    fig, axes = plt.subplots(1, 2, figsize=(14, 5.5))
     fig.patch.set_facecolor("#0d0d0d")
 
+    # ── Left: 7×7 network heatmap ──────────────────────────────────────────
     ax = axes[0]
-    ax.set_facecolor("#111"); ax.tick_params(colors="#555", labelsize=8)
-    for sp in ax.spines.values(): sp.set_color("#222")
-    im = ax.imshow(sal_top, cmap="inferno", aspect="auto", interpolation="nearest")
-    ax.set_title("FC Edge Saliency  (top 5% connections)", color="#bbb", fontsize=10, pad=10)
-    ax.set_xlabel("ROI index", color="#555", fontsize=9)
-    ax.set_ylabel("ROI index", color="#555", fontsize=9)
+    ax.set_facecolor("#111")
+    im = ax.imshow(net_sal, cmap="inferno", aspect="auto", interpolation="nearest")
+    ax.set_title("FC Saliency by Brain Network", color="#bbb", fontsize=11, pad=14, fontweight="bold")
+
+    ax.set_xticks(range(n_nets))
+    ax.set_yticks(range(n_nets))
+    ax.set_xticklabels(_NET_NAMES, rotation=40, ha="right", fontsize=9, color="#ccc")
+    ax.set_yticklabels(_NET_NAMES, fontsize=9, color="#ccc")
+    ax.tick_params(colors="#555", length=0)
+    for sp in ax.spines.values():
+        sp.set_color("#222")
+
+    # Boundary lines between networks
+    for k in range(1, n_nets):
+        ax.axhline(k - 0.5, color="#2a2a2a", lw=1.0)
+        ax.axvline(k - 0.5, color="#2a2a2a", lw=1.0)
+
+    # Annotate each cell with its value
+    vmax = net_sal.max()
+    for i in range(n_nets):
+        for j in range(n_nets):
+            txt_color = "#111" if net_sal[i, j] > 0.6 * vmax else "#555"
+            ax.text(j, i, f"{net_sal[i, j]:.3f}", ha="center", va="center",
+                    fontsize=6.5, color=txt_color)
+
     cb = plt.colorbar(im, ax=ax, fraction=0.046, pad=0.04)
     cb.ax.yaxis.set_tick_params(color="#444", labelsize=7)
     plt.setp(cb.ax.yaxis.get_ticklabels(), color="#555")
+    cb.set_label("Mean |∂p(ASD)/∂FC|", color="#444", fontsize=7.5)
 
+    # ── Right: network importance bar chart ────────────────────────────────
     ax2 = axes[1]
-    ax2.set_facecolor("#111"); ax2.tick_params(colors="#555", labelsize=8)
-    ax2.barh(range(20), roi_imp[top20], color=color, alpha=0.8, edgecolor="none")
-    ax2.set_yticks(range(20))
-    ax2.set_yticklabels([f"ROI {i:03d}" for i in top20], fontsize=8, color="#aaa")
-    ax2.set_xlabel("Cumulative gradient magnitude", color="#555", fontsize=9)
-    ax2.set_title("Top-20 ROIs by Prediction Influence", color="#bbb", fontsize=10, pad=10)
+    ax2.set_facecolor("#111")
+    ax2.tick_params(colors="#555", labelsize=9)
+
+    order = net_imp.argsort()[::-1]
+    bars  = ax2.barh(range(n_nets), net_imp[order],
+                     color=[_NET_COLORS[i] for i in order], alpha=0.88, edgecolor="none", height=0.65)
+    ax2.set_yticks(range(n_nets))
+    ax2.set_yticklabels([_NET_NAMES[i] for i in order], fontsize=9.5, color="#ddd")
+    ax2.set_xlabel("Mean gradient magnitude", color="#555", fontsize=9)
+    ax2.set_title("Network Importance for This Prediction", color="#bbb", fontsize=11, pad=14, fontweight="bold")
     ax2.invert_yaxis()
-    for sp in ["top", "right"]: ax2.spines[sp].set_visible(False)
-    for sp in ["bottom", "left"]: ax2.spines[sp].set_color("#222")
+    for sp in ["top", "right"]:
+        ax2.spines[sp].set_visible(False)
+    for sp in ["bottom", "left"]:
+        ax2.spines[sp].set_color("#222")
+
+    # Value labels on bars
+    x_max = net_imp.max()
+    for bar, val in zip(bars, net_imp[order]):
+        ax2.text(val + x_max * 0.015, bar.get_y() + bar.get_height() / 2,
+                 f"{val:.4f}", va="center", color="#555", fontsize=7.5)
 
     fig.suptitle(
-        f"Gradient Saliency  ·  p(ASD)={p_mean:.3f}  ·  {len(_models)}-model LOSO ensemble",
-        color="#555", fontsize=9, y=1.01,
+        f"Gradient Saliency  ·  p(ASD) = {p_mean:.3f}  ·  {len(_models)}-model LOSO ensemble  ·  CC200 → Yeo-7 networks",
+        color="#444", fontsize=8.5, y=1.02,
     )
     plt.tight_layout()
     buf = io.BytesIO()
-    plt.savefig(buf, format="png", dpi=130, bbox_inches="tight", facecolor="#0d0d0d")
+    plt.savefig(buf, format="png", dpi=140, bbox_inches="tight", facecolor="#0d0d0d")
     plt.close(fig)
     buf.seek(0)
     return Image.open(buf).copy()
@@ -260,6 +307,14 @@ HEADER = """
   <div style="color:#333;font-size:0.72rem;letter-spacing:4px;text-transform:uppercase;margin-top:10px">
     Clinical AI · Resting-state fMRI · Scanner-Site-Invariant Classification
   </div>
+  <div style="color:#444;font-size:0.93rem;margin-top:18px;max-width:720px;line-height:1.75">
+    1 in 44 children is diagnosed with ASD. Today, diagnosis takes years of behavioral observation —
+    no biomarker exists. We trained a scanner-site-invariant GCN on 1,102 subjects across 17 institutions
+    and validated it on <span style="color:#e63946;font-weight:700">529 subjects the model never saw, from sites it was never trained on</span>.
+    The result: <span style="color:#e63946;font-weight:700">AUC 0.7872</span> — not on held-out splits of the same scanner, but
+    across entirely different hospitals. Fine-tuned <span style="color:#f4a261;font-weight:700">Qwen2.5-7B on AMD MI300X</span>
+    then translates raw connectivity patterns into structured clinical language a clinician can act on.
+  </div>
   <div style="display:flex;gap:0;margin-top:28px;border:1px solid #1a1a1a;border-radius:12px;overflow:hidden;max-width:700px">
     <div style="padding:20px 32px;flex:1;border-right:1px solid #1a1a1a;min-width:120px">
       <div style="font-size:2.2rem;font-weight:900;color:#e63946;line-height:1">0.7872</div>