Update app.py

app.py

@@ -136,7 +136,7 @@ CLINICAL_GUIDANCE = (
 
 
 # ============================================================
-# UX HELPERS (NEW)
+# UX HELPERS
 # ============================================================
 def pretty_state(s: str) -> str:
     return {
@@ -147,10 +147,12 @@ def pretty_state(s: str) -> str:
         "N/A": "⚠️ N/A",
     }.get(s, f"⚠️ {s}")
 
+
 def html_escape(s: str) -> str:
     # minimal escaping for safety in HTML blocks
     return (s or "").replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
 
+
 def badge_color_for_state(state: str) -> str:
     # soft visual cue; works in both dark/light
     if state == "TB+":
@@ -181,6 +183,7 @@ class DoubleConv(nn.Module):
 
     def forward(self, x): return self.net(x)
 
+
 class LungUNet(nn.Module):
     def __init__(self):
         super().__init__()
@@ -224,10 +227,12 @@ class TBNet(nn.Module):
 
     def forward(self, x): return self.fc(self.backbone(x)).view(-1)
 
+
 def load_tb_weights(model: nn.Module, ckpt_path: str, device: torch.device):
     sd = torch.load(ckpt_path, map_location=device)
     model.load_state_dict(sd, strict=True)
 
+
 class GradCAM:
     def __init__(self, model: nn.Module, target_layer: nn.Module):
         self.model = model
@@ -271,6 +276,7 @@ def preprocess_for_lung_unet(gray_u8: np.ndarray) -> torch.Tensor:
     g = (g - lo) / (hi - lo + 1e-8)
     return torch.from_numpy(g).unsqueeze(0).unsqueeze(0).float()
 
+
 def tb_training_preprocess(gray_u8: np.ndarray) -> np.ndarray:
     gray = gray_u8.astype(np.float32)
     lo, hi = np.percentile(gray, (1, 99))
@@ -278,16 +284,19 @@ def tb_training_preprocess(gray_u8: np.ndarray) -> np.ndarray:
     gray = (gray - lo) / (hi - lo + 1e-8)
     return gray
 
+
 def laplacian_sharpness(gray_u8: np.ndarray) -> float:
     g = cv2.resize(gray_u8, (512, 512), interpolation=cv2.INTER_AREA)
     g = cv2.GaussianBlur(g, (3, 3), 0)
     return float(cv2.Laplacian(g, cv2.CV_64F).var())
 
+
 def exposure_scores(gray_u8: np.ndarray) -> Tuple[float, float]:
     lo = float((gray_u8 < 10).mean())
     hi = float((gray_u8 > 245).mean())
     return lo, hi
 
+
 def border_fraction(gray_u8: np.ndarray) -> float:
     h, w = gray_u8.shape
     b = max(5, int(0.06 * min(h, w)))
@@ -299,6 +308,7 @@ def border_fraction(gray_u8: np.ndarray) -> float:
     def frac_border(x): return float(((x < 15) | (x > 240)).mean())
     return float(np.mean([frac_border(top), frac_border(bot), frac_border(left), frac_border(right)]))
 
+
 def phone_quality_report(gray_u8: np.ndarray) -> Tuple[float, List[str]]:
     warnings: List[str] = []
     h, w = gray_u8.shape
@@ -334,6 +344,7 @@ def phone_quality_report(gray_u8: np.ndarray) -> Tuple[float, List[str]]:
 
     return float(np.clip(score, 0, 100)), warnings
 
+
 def auto_border_crop(gray_u8: np.ndarray) -> np.ndarray:
     g = gray_u8.copy()
     g_blur = cv2.GaussianBlur(g, (5, 5), 0)
@@ -357,10 +368,12 @@ def auto_border_crop(gray_u8: np.ndarray) -> np.ndarray:
     x2 = min(W, x + w + pad); y2 = min(H, y + h + pad)
     return gray_u8[y1:y2, x1:x2]
 
+
 def apply_clahe(gray_u8: np.ndarray) -> np.ndarray:
     clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
     return clahe.apply(gray_u8)
 
+
 def phone_preprocess(gray_u8: np.ndarray) -> np.ndarray:
     sharp = laplacian_sharpness(gray_u8)
     lo_clip, _hi_clip = exposure_scores(gray_u8)
@@ -377,11 +390,13 @@ def phone_preprocess(gray_u8: np.ndarray) -> np.ndarray:
 
     return g
 
+
 def cam_entropy(cam: np.ndarray) -> float:
     cam = cam.astype(np.float32)
     cam = cam / (cam.sum() + 1e-8)
     return float(-np.sum(cam * np.log(cam + 1e-8)))
 
+
 def detect_diffuse_risk(prob_tb: float, cam_up: np.ndarray, quality_score: float) -> bool:
     if quality_score < 55:
         return False
@@ -390,6 +405,7 @@ def detect_diffuse_risk(prob_tb: float, cam_up: np.ndarray, quality_score: float
     ent = cam_entropy(cam_up)
     return (prob_tb < TBNET_SCREEN_THR) and (ent > 6.5)
 
+
 def confidence_band(prob_tb: float, quality_score: float, diffuse: bool):
     if prob_tb < 0.01 and quality_score >= 45:
         return ("GREEN", "✅ Very low TB signal detected.")
@@ -401,21 +417,24 @@ def confidence_band(prob_tb: float, quality_score: float, diffuse: bool):
         return ("YELLOW", "⚠️ Screening-positive range; review recommended.")
     return ("GREEN", "✅ No strong TB signal detected.")
 
+
 def make_mask_overlay(gray_u8: np.ndarray, mask_u8: np.ndarray) -> np.ndarray:
     base = cv2.cvtColor(gray_u8, cv2.COLOR_GRAY2RGB)
     mask_color = cv2.applyColorMap((mask_u8 * 255).astype(np.uint8), cv2.COLORMAP_JET)
     return cv2.addWeighted(base, 0.75, mask_color, 0.25, 0)
 
+
 def fill_holes(binary_u8: np.ndarray) -> np.ndarray:
     m = (binary_u8 * 255).astype(np.uint8)
     h, w = m.shape
     flood = m.copy()
-    mask = np.zeros((h+2, w+2), np.uint8)
+    mask = np.zeros((h + 2, w + 2), np.uint8)
     cv2.floodFill(flood, mask, (0, 0), 255)
     holes = cv2.bitwise_not(flood)
     filled = cv2.bitwise_or(m, holes)
     return (filled > 0).astype(np.uint8)
 
+
 def keep_top_k_components(binary_u8: np.ndarray, k: int = 2) -> np.ndarray:
     m = (binary_u8 > 0).astype(np.uint8)
     n, labels = cv2.connectedComponents(m)
@@ -431,6 +450,7 @@ def keep_top_k_components(binary_u8: np.ndarray, k: int = 2) -> np.ndarray:
         out[labels == i] = 1
     return out
 
+
 def mask_sanity_warnings(mask_full_u8: np.ndarray) -> List[str]:
     m = (mask_full_u8 > 0).astype(np.uint8)
     n, labels = cv2.connectedComponents(m)
@@ -460,6 +480,7 @@ def mask_sanity_warnings(mask_full_u8: np.ndarray) -> List[str]:
 
     return warns
 
+
 def recommendation_for_band(band: Optional[str]) -> str:
     if band in (None, "YELLOW"):
         return "✅ Recommendation: Radiologist/clinician review is recommended (**indeterminate**)."
@@ -482,6 +503,7 @@ def tbnet_state(tb_prob: float, tb_band: str) -> str:
         return "SCREEN+"
     return "LOW"
 
+
 def radio_state_from_prob(radio_prob: float) -> str:
     if radio_prob >= RADIO_THR_RED:
         return "TB+"
@@ -489,6 +511,7 @@ def radio_state_from_prob(radio_prob: float) -> str:
         return "SCREEN+"
     return "LOW"
 
+
 def build_consensus(
     tb_prob: Optional[float],
     tb_band: Optional[str],
@@ -622,6 +645,7 @@ class ModelBundle:
             self.lung = lung
             self.lung_path = lung_weights
 
+
 BUNDLE = ModelBundle()
 
 
@@ -643,6 +667,7 @@ class RadioMLPHead(nn.Module):
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         return self.net(x).squeeze(1)
 
+
 class RadioBundle:
     def __init__(self):
         self.loaded = False
@@ -691,8 +716,10 @@ class RadioBundle:
         self.device_str = dev_str
         self.loaded = True
 
+
 RADIO_BUNDLE = RadioBundle()
 
+
 def radio_heatmap_from_spatial(spatial_tokens: torch.Tensor, in_h: int, in_w: int, patch_size: int = 16) -> np.ndarray:
     ht = in_h // patch_size
     wt = in_w // patch_size
@@ -703,6 +730,7 @@ def radio_heatmap_from_spatial(spatial_tokens: torch.Tensor, in_h: int, in_w: in
     hm_img = Image.fromarray((hm * 255).astype(np.uint8)).resize((in_w, in_h), resample=Image.BILINEAR)
     return np.array(hm_img, dtype=np.float32) / 255.0
 
+
 def radio_overlay_heatmap(rgb_u8: np.ndarray, heatmap01: np.ndarray, alpha: float = 0.35) -> np.ndarray:
     img = rgb_u8.astype(np.float32) / 255.0
     hm = np.clip(heatmap01, 0, 1).astype(np.float32)
@@ -710,6 +738,7 @@ def radio_overlay_heatmap(rgb_u8: np.ndarray, heatmap01: np.ndarray, alpha: floa
     out[..., 0] = np.clip(out[..., 0] * (1 - alpha) + hm * alpha, 0, 1)
     return (out * 255).astype(np.uint8)
 
+
 @torch.inference_mode()
 def radio_predict_from_arrays(gray_vis_u8: np.ndarray,
                              lung_mask_u8: np.ndarray,
@@ -1076,7 +1105,6 @@ def run_analysis(
         # ------------------------
         # Build descriptive cards (TBNet, RADIO, Consensus)
         # ------------------------
-        # TBNet block values
         tb_prob_line = "N/A (fail-safe)" if out["prob"] is None else f"{out['prob']:.4f}"
         tb_band = out.get("band", "YELLOW")
         tb_label = out.get("pred", "INDETERMINATE")
@@ -1089,12 +1117,10 @@ def run_analysis(
 
         attention = "Diffuse / non-focal" if out.get("diffuse_risk", False) else "Focal / localized"
 
-        # RADIO block values
         radio_primary_line = "N/A" if radio_primary_val is None else f"{radio_primary_val:.4f}"
         radio_raw_line = "N/A" if radio_raw_val is None else f"{radio_raw_val:.4f}"
         radio_masked_line = "Not run" if radio_masked_val is None else f"{radio_masked_val:.4f}"
 
-        # Recommended next step (consensus-aware)
         if consensus_label == "DISAGREE":
             next_step = "✅ Next step: Treat as **indeterminate** → radiologist review + microbiology if clinically suspected."
         elif consensus_label in ("AGREE: TB+", "AGREE: SCREEN+"):
@@ -1104,7 +1130,6 @@ def run_analysis(
         else:
             next_step = "✅ Next step: Correlate clinically; radiologist review recommended if uncertainty or symptoms present."
 
-        # Card containers (HTML)
         state_badge_tb = f"""
 <span style="padding:4px 10px; border-radius:999px; background:{badge_color_for_state(tb_state)}; font-weight:800;">
   {pretty_state(tb_state)}
@@ -1196,7 +1221,6 @@ def run_analysis(
         # Detailed collapsible report (per image)
         # ------------------------
         warn_txt = "\n".join([f"- {w}" for w in out["warnings"]]) if out["warnings"] else "- None"
-        rec_line = recommendation_for_band(out.get("band"))
 
         details_md.append(
             f"""
@@ -1243,7 +1267,7 @@ def run_analysis(
 
 
 # ============================================================
-# UI
+# UI (HF Spaces Welcome Screen + Main App)
 # ============================================================
 def build_ui():
     css = """
@@ -1251,101 +1275,205 @@ def build_ui():
     .subtitle {font-size: 14px; opacity: 0.88; margin-bottom: 14px;}
     .warnbox {border-left: 6px solid #f59e0b; padding: 10px 12px; background: rgba(245,158,11,0.08); border-radius: 10px;}
     .legend {border-left: 6px solid rgba(148,163,184,0.7); padding: 10px 12px; background: rgba(148,163,184,0.08); border-radius: 10px;}
+    .card {border:1px solid rgba(255,255,255,0.12); border-radius:14px; padding:14px; margin:10px 0;}
     """
 
     with gr.Blocks(title="TB X-ray Assistant (TBNet + RADIO)", css=css) as demo:
-        gr.Markdown('<div class="title">TB X-ray Assistant (Auto Lung Mask • Research Use)</div>')
-        gr.Markdown(
-            f"<div class='subtitle'>Auto lung mask → <b>{MODEL_NAME_TBNET}</b> + Grad-CAM • "
-            f"Optional <b>{MODEL_NAME_RADIO}</b> (C-RADIOv4 + heads) • Clear per-model results + consensus</div>"
-        )
-
-        # Disclaimer shown early (so users see it before running)
-        gr.Markdown(
-            "<div class='warnbox'><b>Clinical disclaimer:</b> Decision support only (not diagnostic). "
-            "If TB is clinically suspected, pursue microbiology / CT as appropriate regardless of AI output.</div>"
-        )
 
-        with gr.Row():
-            with gr.Column(scale=1):
-                gr.Markdown("#### Model settings")
-
-                tb_weights = gr.Textbox(label="TBNet weights (.pt)", value=DEFAULT_TB_WEIGHTS)
-                lung_weights = gr.Textbox(label="Lung U-Net weights (.pt)", value=DEFAULT_LUNG_WEIGHTS)
+        # ---------------------------
+        # Welcome screen (shown first)
+        # ---------------------------
+        with gr.Column(visible=True) as welcome_screen:
+            gr.Markdown('<div class="title">Welcome — TB X-ray Assistant (HF Spaces)</div>')
+
+            gr.Markdown(
+                f"""
+<div class="card">
+  <div style="font-size:16px; font-weight:900; margin-bottom:8px;">What this Space does</div>
+  <div style="opacity:0.92;">
+    This app analyzes chest X-rays for <b>TB-like patterns</b> and shows results from:
+    <ul>
+      <li><b>{MODEL_NAME_TBNET}</b> with <b>Grad-CAM</b> explainability</li>
+      <li><b>Auto Lung Mask</b> (Lung U-Net) + a <b>fail-safe</b> to prevent unsafe scoring on bad masks</li>
+      <li>Optional <b>{MODEL_NAME_RADIO}</b> with <b>RAW / MASKED</b> scoring and heatmaps</li>
+      <li>A final <b>consensus</b>: ✅ LOW / ⚠️ INDET / ⚠️ SCREEN+ / 🚩 TB+</li>
+    </ul>
+  </div>
+</div>
 
-                backbone = gr.Dropdown(choices=["efficientnet_b0"], value="efficientnet_b0", label="TBNet backbone")
+<div class="warnbox">
+  <b>Clinical disclaimer:</b> Decision support only (not diagnostic). TB can be subtle (including miliary TB).
+  If TB is clinically suspected, pursue microbiology (CBNAAT/GeneXpert, sputum) and/or CT chest regardless of AI output.
+</div>
 
-                threshold = gr.Slider(
-                    0.01, 0.99, value=TBNET_SCREEN_THR, step=0.01,
-                    label=f"Reference threshold (TBNet screen+) = {TBNET_SCREEN_THR:.2f}"
-                )
+<div class="card">
+  <div style="font-size:16px; font-weight:900; margin-bottom:8px;">Special feature: Phone / WhatsApp Mode</div>
+  <div style="opacity:0.92;">
+    Many users upload:
+    <ul>
+      <li><b>WhatsApp-forwarded X-rays</b> (compressed, low contrast)</li>
+      <li><b>Phone photos</b> of printed films or monitor screens (borders, glare, blur)</li>
+      <li><b>Screenshots</b> with large margins / UI elements</li>
+    </ul>
+    <b>Phone / WhatsApp Mode</b> is designed for these cases. When enabled, it applies:
+    <ul>
+      <li><b>Safe border crop</b> (reduces margins / screenshot framing)</li>
+      <li><b>Conditional CLAHE</b> (boosts local contrast when underexposed / low-detail)</li>
+      <li><b>Quality warnings</b> (blur, over/underexposure, heavy borders) to flag reduced reliability</li>
+    </ul>
+
+    <div style="margin-top:10px; padding:10px 12px; border-left:6px solid rgba(96,165,250,0.9); background: rgba(96,165,250,0.10); border-radius:12px;">
+      <b>Tip:</b> Enable Phone/WhatsApp Mode if your image is a phone photo, WhatsApp-forwarded, or has big borders / low contrast.
+      Keep it OFF for clean digital exports to avoid unnecessary preprocessing.
+    </div>
+  </div>
+</div>
 
-                phone_mode = gr.Checkbox(
-                    value=False,
-                    label="Phone/WhatsApp Mode (safe: conditional crop + conditional CLAHE)"
-                )
+<div class="card">
+  <div style="font-size:16px; font-weight:900; margin-bottom:8px;">Explainability & reliability</div>
+  <ul>
+    <li><b>Grad-CAM</b> (TBNet) highlights regions that influenced the TB score.</li>
+    <li><b>RADIO heatmaps</b> show where the visual model is focusing (RAW and sometimes MASKED).</li>
+    <li><b>Fail-safe</b>: if lung segmentation looks unreliable, TBNet scoring is disabled (shown as indeterminate).</li>
+    <li><b>Quality scoring</b> warns when results may be less reliable (blur, compression, non-standard view).</li>
+  </ul>
+</div>
 
-                use_radio = gr.Checkbox(value=False, label=f"Enable {MODEL_NAME_RADIO}")
-                radio_gate = gr.Slider(
-                    0.10, 0.40, value=RADIO_GATE_DEFAULT, step=0.01,
-                    label="RADIO masked gate (run masked head if lung coverage ≥ gate)"
-                )
+<div class="card">
+  <div style="font-size:16px; font-weight:900; margin-bottom:8px;">How to use</div>
+  <ol>
+    <li>Click <b>Continue</b> to open the interface.</li>
+    <li>Upload one or multiple X-ray images.</li>
+    <li>If your images come from <b>WhatsApp / phone camera / screenshots</b>, enable <b>Phone/WhatsApp Mode</b>.</li>
+    <li>(Optional) Enable <b>{MODEL_NAME_RADIO}</b> for a second independent model + heatmaps.</li>
+    <li>Click <b>Run Analysis</b>.</li>
+  </ol>
+</div>
 
-                gr.Markdown(
-                    "<div class='warnbox'><b>Fail-safe:</b> If lung segmentation is too small or looks unreliable, "
-                    f"{MODEL_NAME_TBNET} scoring is disabled to avoid unsafe outputs.</div>"
-                )
+<div class="card">
+  <div style="font-size:16px; font-weight:900; margin-bottom:8px;">Privacy / processing note (HF Spaces)</div>
+  <div style="opacity:0.92;">
+    Images are processed by this Space runtime. Avoid uploading personally identifiable medical data.
+    Use anonymized images when possible.
+  </div>
+</div>
 
-                gr.Markdown(
-                    f"<div class='subtitle'>Device: <b>{DEVICE}</b> (FORCE_CPU={FORCE_CPU})</div>"
-                )
+<div class="subtitle">
+  Device policy: <b>{DEVICE}</b> (FORCE_CPU={FORCE_CPU})
+</div>
+"""
+            )
+            continue_btn = gr.Button("Continue →", variant="primary")
+
+        # ---------------------------
+        # Main app UI (hidden initially)
+        # ---------------------------
+        with gr.Column(visible=False) as main_app:
+            gr.Markdown('<div class="title">TB X-ray Assistant (Auto Lung Mask • Research Use)</div>')
+            gr.Markdown(
+                f"<div class='subtitle'>Auto lung mask → <b>{MODEL_NAME_TBNET}</b> + Grad-CAM • "
+                f"Optional <b>{MODEL_NAME_RADIO}</b> (C-RADIOv4 + heads) • Clear per-model results + consensus</div>"
+            )
 
-            with gr.Column(scale=2):
-                gr.Markdown("#### Upload images")
-                files = gr.Files(
-                    label="Upload one or multiple X-ray images",
-                    file_types=[".png", ".jpg", ".jpeg", ".bmp"]
-                )
-                run_btn = gr.Button("Run Analysis", variant="primary")
-                status = gr.Textbox(label="Status", value="Ready.", interactive=False)
+            gr.Markdown(
+                "<div class='warnbox'><b>Clinical disclaimer:</b> Decision support only (not diagnostic). "
+                "If TB is clinically suspected, pursue microbiology / CT as appropriate regardless of AI output.</div>"
+            )
 
-                gr.Markdown("""
+            with gr.Row():
+                with gr.Column(scale=1):
+                    gr.Markdown("#### Model settings")
+
+                    tb_weights = gr.Textbox(label="TBNet weights (.pt)", value=DEFAULT_TB_WEIGHTS)
+                    lung_weights = gr.Textbox(label="Lung U-Net weights (.pt)", value=DEFAULT_LUNG_WEIGHTS)
+
+                    backbone = gr.Dropdown(choices=["efficientnet_b0"], value="efficientnet_b0", label="TBNet backbone")
+
+                    threshold = gr.Slider(
+                        0.01, 0.99, value=TBNET_SCREEN_THR, step=0.01,
+                        label=f"Reference threshold (TBNet screen+) = {TBNET_SCREEN_THR:.2f}"
+                    )
+
+                    phone_mode = gr.Checkbox(
+                        value=False,
+                        label="Phone/WhatsApp Mode (safe: conditional crop + conditional CLAHE)"
+                    )
+                    gr.Markdown(
+                        "<div class='subtitle'>Enable for WhatsApp images, phone photos, or screenshots. "
+                        "Leave off for clean digital exports.</div>"
+                    )
+
+                    use_radio = gr.Checkbox(value=False, label=f"Enable {MODEL_NAME_RADIO}")
+                    radio_gate = gr.Slider(
+                        0.10, 0.40, value=RADIO_GATE_DEFAULT, step=0.01,
+                        label="RADIO masked gate (run masked head if lung coverage ≥ gate)"
+                    )
+
+                    gr.Markdown(
+                        "<div class='warnbox'><b>Fail-safe:</b> If lung segmentation is too small or looks unreliable, "
+                        f"{MODEL_NAME_TBNET} scoring is disabled to avoid unsafe outputs.</div>"
+                    )
+
+                    gr.Markdown(
+                        f"<div class='subtitle'>Device: <b>{DEVICE}</b> (FORCE_CPU={FORCE_CPU})</div>"
+                    )
+
+                    back_btn = gr.Button("← Back to Welcome", variant="secondary")
+
+                with gr.Column(scale=2):
+                    gr.Markdown("#### Upload images")
+                    files = gr.Files(
+                        label="Upload one or multiple X-ray images",
+                        file_types=[".png", ".jpg", ".jpeg", ".bmp"]
+                    )
+                    run_btn = gr.Button("Run Analysis", variant="primary")
+                    status = gr.Textbox(label="Status", value="Ready.", interactive=False)
+
+                    gr.Markdown("""
 <div class='legend'><b>Gallery legend:</b><br/>
 1) ORIGINAL &nbsp;•&nbsp; 2) INPUT / PHONE-PROC &nbsp;•&nbsp; 3) Lung mask overlay &nbsp;•&nbsp;
 4) Masked model input &nbsp;•&nbsp; 5) TBNet Grad-CAM &nbsp;•&nbsp; 6) RADIO heatmaps</div>
 """)
 
-        # Summary cards (new)
-        gr.Markdown("#### Summary (per image)")
-        summary = gr.Markdown("Upload images and click <b>Run Analysis</b>.")
-
-        with gr.Row():
-            gallery = gr.Gallery(
-                label="Visual outputs",
-                columns=3,
-                height=560
+            gr.Markdown("#### Summary (per image)")
+            summary = gr.Markdown("Upload images and click <b>Run Analysis</b>.")
+            gallery = gr.Gallery(label="Visual outputs", columns=3, height=560)
+
+            with gr.Row():
+                with gr.Column(scale=1):
+                    disclaimer_box = gr.Markdown(CLINICAL_DISCLAIMER)
+                with gr.Column(scale=2):
+                    gr.Markdown("#### Detailed report (expand per image)")
+                    details = gr.Markdown("")
+
+            run_btn.click(
+                fn=run_analysis,
+                inputs=[
+                    files,
+                    tb_weights,
+                    lung_weights,
+                    backbone,
+                    threshold,
+                    phone_mode,
+                    use_radio,
+                    radio_gate,
+                ],
+                outputs=[summary, gallery, details, disclaimer_box, status]
             )
 
-        with gr.Row():
-            with gr.Column(scale=1):
-                disclaimer_box = gr.Markdown(CLINICAL_DISCLAIMER)
-            with gr.Column(scale=2):
-                gr.Markdown("#### Detailed report (expand per image)")
-                details = gr.Markdown("")
-
-        run_btn.click(
-            fn=run_analysis,
-            inputs=[
-                files,
-                tb_weights,
-                lung_weights,
-                backbone,
-                threshold,
-                phone_mode,
-                use_radio,
-                radio_gate,
-            ],
-            outputs=[summary, gallery, details, disclaimer_box, status]
+        # ---------------------------
+        # Transitions
+        # ---------------------------
+        continue_btn.click(
+            fn=lambda: (gr.update(visible=False), gr.update(visible=True)),
+            inputs=[],
+            outputs=[welcome_screen, main_app],
+        )
+
+        back_btn.click(
+            fn=lambda: (gr.update(visible=True), gr.update(visible=False)),
+            inputs=[],
+            outputs=[welcome_screen, main_app],
         )
 
     return demo
@@ -1353,4 +1481,6 @@ def build_ui():
 
 if __name__ == "__main__":
     demo = build_ui()
-    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)
+    # HF Spaces: let the platform manage host/port. queue() helps stability for longer runs.
+    demo.queue(concurrency_count=1)
+    demo.launch(show_error=True)