# SKIN CANCER DETECTION - VISUAL DASHBOARD # Author: Jayanth # # Flow: # 1) (Optional) Upload .keras/.h5 model -> real predictions if provided # 2) Upload 3–8 lesion images (JPG/PNG) # 3) Outputs: # - Cancer / Not Cancer + probabilities # - Dark gallery (aligned) # - Top-risk strip # - Summary chart # - CSV # - ZIP + manual Download button (no auto-download) !pip -q install tensorflow matplotlib import os, zipfile import numpy as np import pandas as pd import matplotlib.pyplot as plt import matplotlib.patheffects as pe import tensorflow as tf from google.colab import files from matplotlib.patches import Rectangle, FancyBboxPatch from datetime import datetime from IPython.display import HTML, display # ============== CONFIG ============== TITLE = "Skin Cancer Detection - Results" AUTHOR = "Jayanth" IMG_SIZE = 224 MALIGNANT_CUT = 0.50 MAX_IMAGES = 8 BG_COLOR = "#0f1116" PANEL_COLOR = "#141824" # ==================================== plt.rcParams.update({ "figure.facecolor": BG_COLOR, "axes.facecolor": BG_COLOR, "axes.edgecolor": "#ffffff", "text.color": "#ffffff", "savefig.facecolor": BG_COLOR, }) # ---------- Basic helpers ---------- def conf_color(p): return (0.88, 0.12, 0.15) if p >= MALIGNANT_CUT else (0.10, 0.75, 0.35) def fmt_pct(p): return f"{p*100:.1f}%" def draw_header(fig): ax = fig.add_axes([0.0, 0.93, 1.0, 0.07]) ax.set_axis_off() ax.add_patch(Rectangle((0,0),1,1, transform=ax.transAxes, color=PANEL_COLOR, alpha=1.0)) ts = datetime.now().strftime("%Y-%m-%d %H:%M") ax.text(0.02, 0.5, TITLE, fontsize=19, fontweight="bold", va="center", ha="left", color="white") ax.text(0.98, 0.5, f"{AUTHOR} • {ts}", fontsize=10, va="center", ha="right", color="#d0d0d0") def draw_footer(fig, text="For academic/demo use only. Not medical advice."): ax = fig.add_axes([0.0, 0.0, 1.0, 0.035]) ax.set_axis_off() ax.add_patch(Rectangle((0,0),1,1, transform=ax.transAxes, color=PANEL_COLOR, alpha=1.0)) ax.text(0.5, 0.5, text, fontsize=8.5, va="center", ha="center", color="#aaaaaa") def draw_legend(fig): ax = fig.add_axes([0.02, 0.035, 0.22, 0.06]) ax.set_axis_off() ax.add_patch(Rectangle((0,0),1,1, transform=ax.transAxes, color="black", alpha=0.35)) ax.text(0.05, 0.62, "Legend", fontsize=9.5, fontweight="bold", va="center", ha="left", color="white") ax.add_patch(Rectangle((0.05,0.18),0.08,0.24, transform=ax.transAxes, color=(0.10, 0.75, 0.35))) ax.text(0.15, 0.30, "Not Cancer", fontsize=8.5, va="center", ha="left", color="white") ax.add_patch(Rectangle((0.55,0.18),0.08,0.24, transform=ax.transAxes, color=(0.88, 0.12, 0.15))) ax.text(0.65, 0.30, "Cancer", fontsize=8.5, va="center", ha="left", color="white") def draw_card(ax, image, prob, name): ax.set_axis_off() ax.imshow(image) color = conf_color(prob) # Card border ax.add_patch(FancyBboxPatch( (0,0), 1, 1, transform=ax.transAxes, boxstyle="round,pad=0.012,rounding_size=0.032", fill=False, lw=3.5, edgecolor=color )) # Bottom info band band_h = 0.22 ax.add_patch(Rectangle((0,0),1,band_h, transform=ax.transAxes, color="black", alpha=0.70, lw=0)) label = "Cancer" if prob >= MALIGNANT_CUT else "Not Cancer" ax.text(0.03, 0.03, f"{label} • {fmt_pct(prob)}", transform=ax.transAxes, fontsize=12, fontweight="bold", va="bottom", ha="left", color="white", path_effects=[pe.withStroke(linewidth=2, foreground="black", alpha=0.6)]) # Filename ax.text(0.97, 0.03, name[:26], transform=ax.transAxes, fontsize=8, va="bottom", ha="right", color="#e0e0e0", alpha=0.95) # Confidence bar bar_y = 0.03 + band_h * 0.45 ax.add_patch(Rectangle((0.03, bar_y), 0.94, band_h*0.18, transform=ax.transAxes, fill=False, lw=1.3, edgecolor="white")) ax.add_patch(Rectangle((0.03, bar_y), 0.94 * float(np.clip(prob, 0, 1)), band_h*0.18, transform=ax.transAxes, color=color, alpha=0.95)) # Ticks for t, txt in zip([0, 0.5, 1.0], ["0%", "50%", "100%"]): ax.text(0.03 + 0.94*t, bar_y + band_h*0.23, txt, transform=ax.transAxes, fontsize=7, ha="center", va="bottom", color="white", alpha=0.9) # Benign/Malignant text ax.text(0.03, bar_y - 0.07, f"Benign: {fmt_pct(1-prob)}", transform=ax.transAxes, fontsize=8, color="#b8f2c8" if prob < MALIGNANT_CUT else "#aaaaaa") ax.text(0.97, bar_y - 0.07, f"Malignant: {fmt_pct(prob)}", transform=ax.transAxes, fontsize=8, ha="right", color="#ffb3c1" if prob >= MALIGNANT_CUT else "#aaaaaa") # ---------- Optional Grad-CAM (best-effort) ---------- def try_build_gradcam(model): try: backbone = model.layers[0] final = model.layers[-1] inp = tf.keras.Input(shape=(IMG_SIZE, IMG_SIZE, 3)) feat = backbone(inp, training=False) pooled = tf.keras.layers.GlobalAveragePooling2D()(feat) out = final(pooled) grad_model = tf.keras.Model(inp, [feat, out]) def gradcam(x_batch): with tf.GradientTape() as tape: conv_out, preds = grad_model(x_batch, training=False) loss = preds[:, 0] grads = tape.gradient(loss, conv_out) w = tf.reduce_mean(grads, axis=(1,2), keepdims=True) cam = tf.reduce_sum(w * conv_out, axis=-1) cam = tf.nn.relu(cam) cam = cam / (tf.reduce_max(cam, axis=(1,2), keepdims=True) + 1e-8) cam = tf.image.resize(cam[..., None], (IMG_SIZE, IMG_SIZE))[:, :, :, 0] return cam.numpy() print("✅ Grad-CAM enabled.") return gradcam except Exception: print("ℹ️ Grad-CAM not enabled (model structure not compatible).") return None # ---------- Layouts ---------- def plot_gallery(images, probs, names, save_path): n = min(len(images), MAX_IMAGES) if n == 0: return cols = 2 rows = (n + cols - 1) // cols fig_h = rows * 4.8 + 2.8 fig_w = cols * 5.3 fig = plt.figure(figsize=(fig_w, fig_h)) fig.patch.set_facecolor(BG_COLOR) draw_header(fig) draw_footer(fig) draw_legend(fig) top = 0.90 bottom = 0.08 grid_h = top - bottom cell_h = grid_h / rows cell_w = (1.0 - 0.10) / cols idx = 0 for r in range(rows): for c in range(cols): if idx >= n: break x0 = 0.05 + c * cell_w y0 = top - (r+1)*cell_h ax = fig.add_axes([x0, y0+0.02, cell_w-0.02, cell_h-0.04]) draw_card(ax, images[idx], probs[idx], names[idx]) idx += 1 plt.savefig(save_path, dpi=230, bbox_inches="tight") plt.show() print(f"✅ Gallery saved -> {save_path}") def plot_risk_strip(images, probs, names, save_path): if len(images) == 0: return order = np.argsort(probs)[::-1][:min(3, len(images))] fig = plt.figure(figsize=(9, 2.8)) fig.patch.set_facecolor(BG_COLOR) for i, idx in enumerate(order): ax = fig.add_subplot(1, len(order), i+1) ax.set_axis_off() ax.imshow(images[idx]) ax.add_patch(Rectangle((0,0),1,0.28, transform=ax.transAxes, color="black", alpha=0.78, lw=0)) ax.text(0.03, 0.04, f"#{i+1} {names[idx][:20]}", transform=ax.transAxes, fontsize=8.5, va="bottom", ha="left", color="white", path_effects=[pe.withStroke(linewidth=1.2, foreground="black", alpha=0.7)]) ax.text(0.97, 0.04, f"Malignant: {fmt_pct(probs[idx])}", transform=ax.transAxes, fontsize=9.5, va="bottom", ha="right", color="white", fontweight="bold") ax.add_patch(FancyBboxPatch( (0,0),1,1, transform=ax.transAxes, boxstyle="round,pad=0.01,rounding_size=0.03", fill=False, lw=3, edgecolor=conf_color(probs[idx]) )) plt.tight_layout() plt.savefig(save_path, dpi=220, bbox_inches="tight") plt.show() print(f"✅ Risk strip saved -> {save_path}") def plot_summary(probs, save_path): if len(probs) == 0: return probs = np.array(probs) preds = (probs >= MALIGNANT_CUT).astype(int) n_cancer = int(preds.sum()) n_not = len(preds) - n_cancer avg_m = probs.mean() * 100 fig = plt.figure(figsize=(8.5,3.6)) fig.patch.set_facecolor(BG_COLOR) ax1 = fig.add_subplot(1,2,1) ax1.set_title("Prediction Breakdown", fontsize=10.5, fontweight="bold", color="white") ax1.pie( [n_not, n_cancer], labels=["Not Cancer", "Cancer"], autopct="%1.0f%%", startangle=90, colors=[(0.10,0.75,0.35),(0.88,0.12,0.15)], textprops={"color":"white", "fontsize":8} ) ax1.set_aspect("equal") ax2 = fig.add_subplot(1,2,2) ax2.set_title(f"Malignant Probability (avg: {avg_m:.1f}%)", fontsize=10.5, fontweight="bold", color="white") ax2.hist(probs*100, bins=8, edgecolor="white") ax2.set_xlabel("% malignant", fontsize=8.5) ax2.set_ylabel("count", fontsize=8.5) ax2.tick_params(colors="white", labelsize=8) for s in ax2.spines.values(): s.set_color("white") plt.tight_layout() plt.savefig(save_path, dpi=220, bbox_inches="tight") plt.show() print(f"✅ Summary saved -> {save_path}") # ---------- 1) OPTIONAL: upload model ---------- model = None gradcam_fn = None print("🧠 Optional: Upload your trained model (.keras / .h5).") print(" - If you don't upload or upload something else, demo mode will show UI with fake but stable probabilities.\n") try: up_model = files.upload() except Exception: up_model = {} model_files = [n for n in up_model if n.lower().endswith((".keras",".h5"))] if up_model else [] if model_files: mname = model_files[0] with open(f"/content/{mname}", "wb") as f: f.write(up_model[mname]) model = tf.keras.models.load_model(f"/content/{mname}") print(f"✅ Model loaded: {mname}") gradcam_fn = try_build_gradcam(model) elif up_model: # Something uploaded, but not a model print("ℹ️ Uploaded file is not .keras/.h5. Ignoring and using demo mode.") else: print("ℹ️ No model uploaded. Running in demo mode (UI showcase).") # ---------- 2) Upload images ---------- print("\n📤 Upload 3–8 lesion images (JPG/PNG):") uploaded = files.upload() if not uploaded: raise SystemExit("No images uploaded.") image_names = [n for n in uploaded if n.lower().endswith((".jpg",".jpeg",".png"))] if not image_names: raise SystemExit("No JPG/PNG images in upload.") image_names = image_names[:MAX_IMAGES] # ---------- 3) Predict ---------- images = [] probs = [] rows = [] for name in image_names: data = uploaded[name] img = tf.io.decode_image(data, channels=3, expand_animations=False) img = tf.image.resize(img, [IMG_SIZE, IMG_SIZE]) x = tf.cast(img, tf.float32) / 255.0 if model is not None: # real model prediction pred = model(x[None, ...], training=False).numpy() p = float(np.clip(pred.ravel()[0], 0.0, 1.0)) else: # demo: deterministic, decent-looking probability from filename seed = abs(hash(name)) % (2**32) rng = np.random.default_rng(seed) p = float(np.clip(rng.normal(0.55, 0.22), 0.05, 0.97)) images.append(x.numpy()) probs.append(p) label = "Cancer" if p >= MALIGNANT_CUT else "Not Cancer" rows.append((name, label, p*100.0, (1.0-p)*100.0)) print(f"{name}: {label} ({p*100:.1f}%)") # ---------- 4) Visualizations & Export ---------- os.makedirs("/content/cards", exist_ok=True) gallery_path = "/content/pred_gallery.png" risk_path = "/content/risk_strip.png" summary_path = "/content/summary_panel.png" csv_path = "/content/predictions.csv" zip_path = "/content/skin_cancer_outputs.zip" plot_gallery(images, probs, image_names, gallery_path) plot_risk_strip(images, probs, image_names, risk_path) plot_summary(probs, summary_path) df = pd.DataFrame(rows, columns=["filename", "prediction", "malignant_%", "benign_%"]) df.to_csv(csv_path, index=False) print("\n📋 Prediction Table:") print(df.to_string(index=False)) print(f"\n✅ CSV saved -> {csv_path}") # Save individual cards for i, (img, p, name) in enumerate(zip(images, probs, image_names), start=1): fig = plt.figure(figsize=(5,5)) fig.patch.set_facecolor(BG_COLOR) ax = fig.add_axes([0,0,1,1]); ax.set_axis_off() draw_card(ax, img, p, name) out = f"/content/cards/card_{i:02d}.png" plt.savefig(out, dpi=230, bbox_inches="tight") plt.close(fig) # Build ZIP (no auto-download) with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as z: for src, arc in [ (gallery_path, "gallery.png"), (risk_path, "risk_strip.png"), (summary_path, "summary_panel.png"), (csv_path, "predictions.csv"), ]: if os.path.exists(src): z.write(src, arcname=arc) for f in sorted(os.listdir("/content/cards")): full = os.path.join("/content/cards", f) if os.path.isfile(full): z.write(full, arcname=f"cards/{f}") print(f"\n📦 ZIP ready at: {zip_path}") # Manual download button (click -> downloads ZIP, nothing auto) zip_name = os.path.basename(zip_path) display(HTML(f"""
"""))