Upload app.py with huggingface_hub

app.py

@@ -88,10 +88,23 @@ def _compute_saliency(bw_t, adj_t, models):
     sal = np.mean(maps, axis=0)
     return (sal + sal.T) / 2
 
+# Approximate MNI centroids for each CC200 network (mm), used for 3D brain view
+_NET_MNI = np.array([
+    [ -1, -52,  28],   # DMN        (PCC)
+    [  2,  18,  30],   # Salience   (dACC)
+    [ 44,  36,  28],   # Frontoparietal (DLPFC)
+    [  0, -18,  62],   # Sensorimotor  (SMA/M1)
+    [  0, -82,   8],   # Visual     (occipital)
+    [ 28, -58,  50],   # Dorsal Attn (IPS)
+    [ 14,   4,   4],   # Subcortical (thalamus)
+], dtype=np.float32)
+
 def _saliency_figure(sal, p_mean):
     import matplotlib
     matplotlib.use("Agg")
     import matplotlib.pyplot as plt
+    from mpl_toolkits.mplot3d import Axes3D  # noqa: F401
+    from mpl_toolkits.mplot3d.art3d import Line3DCollection
     from PIL import Image
 
     n_nets = len(_NET_NAMES)
@@ -108,8 +121,13 @@ def _saliency_figure(sal, p_mean):
         for s, e in zip(_NET_BOUNDS[:-1], _NET_BOUNDS[1:])
     ])
 
-    fig, axes = plt.subplots(1, 2, figsize=(14, 5.5))
-    fig.patch.set_facecolor("#0d0d0d")
+    fig = plt.figure(figsize=(18, 5.5))
+    fig.patch.set_facecolor("#0e1015")
+    axes = [
+        fig.add_subplot(1, 3, 1),
+        fig.add_subplot(1, 3, 2),
+        fig.add_subplot(1, 3, 3, projection="3d"),
+    ]
 
     # ── Left: 7×7 network heatmap ──────────────────────────────────────────
     ax = axes[0]
@@ -192,6 +210,45 @@ def _saliency_figure(sal, p_mean):
         ax2.text(val + x_max * 0.015, bar.get_y() + bar.get_height() / 2,
                  f"{val:.4f}", va="center", color="#555", fontsize=7.5)
 
+    # ── 3D Brain Surface — top connections ────────────────────────────────────
+    ax3 = axes[2]
+    ax3.set_facecolor("#0e1015")
+    ax3.grid(False)
+    ax3.set_axis_off()
+    ax3.set_title("Top Connections · 3D Brain", color="#bbb", fontsize=11, pad=4, fontweight="bold")
+
+    # Transparent brain ellipsoid wireframe (MNI space approx)
+    u = np.linspace(0, 2 * np.pi, 32)
+    v = np.linspace(0, np.pi, 20)
+    ex = 68 * np.outer(np.cos(u), np.sin(v))
+    ey = 85 * np.outer(np.sin(u), np.sin(v)) - 10
+    ez = 60 * np.outer(np.ones_like(u), np.cos(v)) + 28
+    ax3.plot_wireframe(ex, ey, ez, color="#252a35", linewidth=0.25, alpha=0.45, zorder=0)
+
+    # Network nodes — size proportional to importance
+    imp_norm = (net_imp - net_imp.min()) / (net_imp.max() - net_imp.min() + 1e-9)
+    for k, (name, color) in enumerate(zip(_NET_NAMES, _NET_COLORS)):
+        x, y, z = _NET_MNI[k]
+        size = 60 + imp_norm[k] * 260
+        ax3.scatter([x], [y], [z], c=color, s=size, zorder=5,
+                    edgecolors="#ffffff", linewidths=0.5, alpha=0.92)
+        ax3.text(x, y, z + 7, name, fontsize=5.5, color=color,
+                 ha="center", va="bottom", fontweight="600", zorder=6)
+
+    # Draw top-5 inter-network connections as lines, thickness ∝ saliency
+    sal_vals = [s for s, _, _ in edge_scores[:5]]
+    sal_min, sal_max = min(sal_vals), max(sal_vals) + 1e-9
+    for rank, (score, ni, nj) in enumerate(edge_scores[:5]):
+        p1, p2 = _NET_MNI[ni], _NET_MNI[nj]
+        lw   = 0.8 + 2.5 * (score - sal_min) / (sal_max - sal_min)
+        alph = 0.5 + 0.45 * (score - sal_min) / (sal_max - sal_min)
+        clr  = "#fb923c" if rank == 0 else "#f4f4f5"
+        ax3.plot([p1[0], p2[0]], [p1[1], p2[1]], [p1[2], p2[2]],
+                 color=clr, linewidth=lw, alpha=alph, zorder=4)
+
+    ax3.view_init(elev=22, azim=-65)
+    ax3.set_box_aspect([1.2, 1.4, 1.0])
+
     fig.suptitle(
         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,