{"repo": "allanotieno254/-breast-cancer-ultrasound-classifier-enhanced", "files": {"check.py": {"1c4d3a54fba097ec96fea00114fa72485d8630a7": {"t": 1785745427.473821, "n": 13174, "h": "53da1fdb056298b3", "c": "# # app.py (merged, schema-migrating version — fixes \"no such column\" errors)\n# import streamlit as st\n# from PIL import Image\n# from img_classification import (\n#     teachable_machine_classification,\n#     grad_cam_for_pil,\n#     is_image_blurry,\n#     prepare_result_record,\n#     insert_record,\n#     get_model_metadata,\n# )\n# import base64\n# import os\n# import requests\n# from dotenv import load_dotenv\n# from io import BytesIO\n# import json\n# import matplotlib.pyplot as plt\n# import tempfile\n# import sqlite3\n# import pandas as pd\n\n# load_dotenv()\n\n# DEEPSEEK_API_KEY = os.getenv(\"DEEPSEEK_API_KEY\")\n# DEEPSEEK_MODEL = os.getenv(\"DEEPSEEK_MODEL\", \"deepseek-chat\")\n# DEEPSEEK_BASE = os.getenv(\"DEEPSEEK_BASE_URL\", \"https://api.deepseek.com/v1\")\n\n# st.set_page_config(page_title=\"Breast Cancer Classifier\", layout=\"wide\")\n\n# DB_PATH = os.path.join(os.path.dirname(__file__), \"predictions.db\")\n\n# # --------------------------------------------------------------------\n# # DATABASE MIGRATION / ENSURE SCHEMA\n# # --------------------------------------------------------------------\n# REQUIRED_COLUMNS = {\n#     \"id\": \"INTEGER PRIMARY KEY AUTOINCREMENT\",\n#     \"timestamp\": \"TEXT\",\n#     \"patient_name\": \"TEXT\",\n#     \"patient_age\": \"INTEGER\",\n#     \"patient_gender\": \"TEXT\",\n#     \"notes\": \"TEXT\",\n#     \"image_paths\": \"TEXT\",\n#     \"prediction\": \"TEXT\",\n#     \"probs\": \"TEXT\",\n#     \"explanation\": \"TEXT\",\n#     \"model_metadata\": \"TEXT\",\n#     \"psno\": \"TEXT\"\n# }\n\n# def ensure_table_and_columns():\n#     \"\"\"\n#     Ensures `predictions` table exists and contains all REQUIRED_COLUMNS.\n#     If table exists but misses columns, add them via ALTER TABLE.\n#     \"\"\"\n#     conn = sqlite3.connect(DB_PATH)\n#     c = conn.cursor()\n\n#     # If table doesn't exist, create with full schema\n#     c.execute(\"SELECT name FROM sqlite_master WHERE type='table' AND name='predictions'\")\n#     if not c.fetchone():\n#         # build create statement using REQUIRED_COLUMNS in a deterministic order\n#         cols_sql = \",\\n    \".join([f\"{col} {REQUIRED_COLUMNS[col]}\" for col in REQUIRED_COLUMNS])\n#         create_sql = f\"CREATE TABLE predictions (\\n    {cols_sql}\\n)\"\n#         c.execute(create_sql)\n#         conn.commit()\n#         conn.close()\n#         return\n\n#     # Table exists: inspect columns\n#     c.execute(\"PRAGMA table_info(predictions)\")\n#     existing = [row[1] for row in c.fetchall()]  # row format: (cid, name, type, notnull, dflt_value, pk)\n#     # Add any missing columns\n#     for col, coldef in REQUIRED_COLUMNS.items():\n#         if col not in existing:\n#             try:\n#                 # For PK we skip altering since it's only for creation; but if id absent, we still add a column id (non-PK)\n#                 if col == \"id\":\n#                     # If id missing, add a plain INTEGER column (can't make it PK on alter)\n#                     c.execute(f\"ALTER TABLE predictions ADD COLUMN {col} INTEGER\")\n#                 else:\n#                     c.execute(f\"ALTER TABLE predictions ADD COLUMN {col} {coldef}\")\n#             except Exception:\n#                 # best-effort: ignore failures (e.g., column already added by race condition)\n#                 pass\n\n#     conn.commit()\n#     conn.close()\n\n# # ensure schema before anything\n# ensure_table_and_columns()\n\n# # --------------------------------------------------------------------\n# # SIDEBAR PATIENT INFO (includes PSNO)\n# # --------------------------------------------------------------------\n# st.sidebar.header(\"Patient Details\")\n# with st.sidebar.form(\"patient_form\"):\n#     psno = st.text_input(\"Patient PSNO (unique ID)\")\n#     name = st.text_input(\"Patient name\")\n#     age = st.number_input(\"Age\", min_value=0, max_value=130, value=40)\n#     gender = st.selectbox(\"Gender\", [\"Female\", \"Male\", \"Other\"])\n#     symptoms = st.text_area(\"Clinical notes (symptoms)\")\n#     submit_patient = st.form_submit_button(\"Save patient info\")\n\n# if submit_patient:\n#     if not psno:\n#         st.sidebar.error(\"PSNO is required.\")\n#     else:\n#         st.sidebar.success(\"Patient info saved (temporary)\")\n\n# # --------------------------------------------------------------------\n# # MODEL METADATA (unchanged)\n# # --------------------------------------------------------------------\n# st.sidebar.header(\"Model Info\")\n# meta = get_model_metadata()\n# st.sidebar.write(f\"Model file: {os.path.basename(meta['model_path'])}\")\n# st.sidebar.write(f\"Input size: {meta['input_size']}\")\n# st.sidebar.write(f\"Labels: {meta['labels']}\")\n# st.sidebar.write(f\"Loaded layers: {len(meta['keras_model_layers'])} layers\")\n\n# # --------------------------------------------------------------------\n# # MAIN: Upload, classify, grad-cam, LLM, insert\n# # --------------------------------------------------------------------\n# st.title(\"🩺 Breast Cancer Ultrasound Classifier — Enhanced\")\n# st.markdown(\"Upload ultrasound images for automated analysis, Grad-CAM, explanation, and DB saving.\")\n\n# uploaded_files = st.file_uploader(\"Upload image(s)\", accept_multiple_files=True, type=[\"jpg\",\"jpeg\",\"png\"])\n# exp_mode = st.radio(\"Explanation mode\", (\"Doctor\", \"Patient (plain language)\"))\n\n# if uploaded_files:\n#     results = []\n#     temp_dir = tempfile.mkdtemp(prefix=\"bcapp2_\")\n\n#     for uploaded in uploaded_files:\n#         try:\n#             image = Image.open(uploaded).convert(\"RGB\")\n#         except Exception:\n#             st.error(f\"Could not open {uploaded.name}\")\n#             continue\n\n#         img_path = os.path.join(temp_dir, uploaded.name)\n#         image.save(img_path)\n\n#         blurry, lap_var = is_image_blurry(image)\n#         if blurry:\n#             st.warning(f\"{uploaded.name} appears blurry (Laplacian variance={lap_var:.1f}).\")\n\n#         pred_idx, probs = teachable_machine_classification(image)\n#         class_names = [\"Malignant\", \"Benign\"]\n#         pred_label = class_names[pred_idx]\n\n#         # GRAD-CAM\n#         try:\n#             overlay, heatmap = grad_cam_for_pil(image)\n#             grad_path = os.path.join(temp_dir, f\"gradcam_{uploaded.name}\")\n#             overlay.save(grad_path)\n#         except Exception:\n#             grad_path = None\n\n#         # LLM (keep behavior from app.py2)\n#         explanation = \"LLM disabled (no API key configured).\"\n#         if DEEPSEEK_API_KEY:\n#             headers = {\"Authorization\": f\"Bearer {DEEPSEEK_API_KEY}\", \"Content-Type\": \"application/json\"}\n#             payload = {\n#                 \"model\": DEEPSEEK_MODEL,\n#                 \"messages\": [\n#                     {\"role\": \"system\", \"content\": \"You are a medical assistant.\"},\n#                     {\"role\": \"user\", \"content\": (\n#                         f\"A breast ultrasound image was analyzed by a machine learning model.\\n\"\n#                         f\"Prediction: {pred_label}\\n\"\n#                         f\"Probabilities: Malignant {probs[0]*100:.2f}%, Benign {probs[1]*100:.2f}%.\\n\"\n#                         f\"Patient name: {name}\\nAge: {age}\\nGender: {gender}\\nNotes: {symptoms}\\n\"\n#                         f\"Mode: {exp_mode}\"\n#                     )}\n#                 ],\n#                 \"max_tokens\": 700,\n#                 \"temperature\": 0.2\n#             }\n#             try:\n#                 r = requests.post(f\"{DEEPSEEK_BASE}/chat/completions\", headers=headers, json=payload, timeout=30)\n#                 if r.status_code == 200:\n#                     data = r.json()\n#                     explanation = data[\"choices\"][0][\"message\"][\"content\"].strip()\n#                 else:\n#                     explanation = f\"DeepSeek error: {r.status_code} {r.text}\"\n#             except Exception as e:\n#                 explanation = f\"DeepSeek API call failed: {e}\"\n\n#         # Recommendation logic (same)\n#         recommendation = \"\"\n#         if probs[0] >= 0.70:\n#             recommendation = \"High suspicion for malignancy — urgent follow-up required.\"\n#         elif probs[1] >= 0.90:\n#             recommendation = \"High confidence benign — routine follow-up.\"\n#         elif max(probs) < 0.55:\n#             recommendation = \"Low confidence — further imaging advised.\"\n\n#         # Prepare record using original prepare_result_record (keeps compatibility)\n#         patient_info = {\"name\": name, \"age\": age, \"gender\": gender, \"notes\": symptoms}\n#         try:\n#             rec = prepare_result_record(patient_info, [img_path], pred_label, probs, explanation)\n#             # rec has fields: timestamp, patient_name, patient_age, patient_gender, notes, image_paths,\n#             #                 prediction, probs, explanation, model_metadata\n#             # Add psno into record so insert_record will store it in the DB's psno column (if insert_record supports it)\n#             rec[\"psno\"] = psno\n#             # Use the original insert_record() from img_classification.py\n#             insert_record(rec)\n#         except Exception as e:\n#             st.error(f\"DB insert error: {e}\")\n\n#         # collect for display\n#         results.append({\n#             \"filename\": uploaded.name,\n#             \"image_path\": img_path,\n#             \"gradcam_path\": grad_path,\n#             \"prediction\": pred_label,\n#             \"probs\": [float(probs[0]), float(probs[1])],\n#             \"explanation\": explanation,\n#             \"recommendation\": recommendation\n#         })\n\n#     # Display results\n#     st.markdown(\"### Results\")\n#     for r in results:\n#         st.markdown(f\"**{r['filename']}** — Prediction: **{r['prediction']}**\")\n#         st.write(f\"Malignant: {r['probs'][0]*100:.2f}% | Benign: {r['probs'][1]*100:.2f}%\")\n#         col1, col2 = st.columns(2)\n#         with col1:\n#             st.image(r[\"image_path\"], caption=\"Original\")\n#         with col2:\n#             if r[\"gradcam_path\"]:\n#                 st.image(r[\"gradcam_path\"], caption=\"Grad-CAM\")\n#         st.info(r[\"recommendation\"])\n#         with st.expander(\"Explanation\"):\n#             st.write(r[\"explanation\"])\n#         fig, ax = plt.subplots(figsize=(4,1.2))\n#         ax.barh([\"Malignant\",\"Benign\"], [r['probs'][0]*100, r['probs'][1]*100])\n#         ax.set_xlim(0,100)\n#         st.pyplot(fig)\n#         st.markdown(\"---\")\n\n# # --------------------------------------------------------------------\n# # DATABASE MANAGEMENT UI (replaces PDF export)\n# # --------------------------------------------------------------------\n# st.header(\"📦 Database Management\")\n\n# # Download CSV\n# if st.button(\"Download database as CSV\"):\n#     ensure_table_and_columns = None  # placeholder to avoid linter confusion (we used ensure_table_and_columns earlier)\n#     # ensure schema and read\n#     ensure_table_and_columns = None\n#     # Re-run the ensure (safe to call)\n#     ensure_table_and_columns = lambda: ensure_table_and_columns  # no-op\n#     # We'll just make sure the DB exists and then read\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         conn.close()\n#         df = pd.DataFrame()\n#     conn.close()\n\n#     if df.empty:\n#         st.info(\"Database is empty.\")\n#     else:\n#         csv = df.to_csv(index=False).encode()\n#         b64 = base64.b64encode(csv).decode()\n#         st.markdown(f'<a href=\"data:file/csv;base64,{b64}\" download=\"predictions.csv\">Download CSV</a>', unsafe_allow_html=True)\n\n# # Download Excel\n# if st.button(\"Download database as Excel\"):\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         conn.close()\n#         df = pd.DataFrame()\n#     conn.close()\n\n#     if df.empty:\n#         st.info(\"Database is empty.\")\n#     else:\n#         excel_path = os.path.join(tempfile.gettempdir(), \"predictions.xlsx\")\n#         df.to_excel(excel_path, index=False)\n#         with open(excel_path, \"rb\") as f:\n#             excel_bytes = f.read()\n#             b64 = base64.b64encode(excel_bytes).decode()\n#             st.markdown(f'<a href=\"data:application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;base64,{b64}\" download=\"predictions.xlsx\">Download Excel</a>', unsafe_allow_html=True)\n\n# # Clear DB\n# if st.button(\"Clear entire database\"):\n#     if os.path.exists(DB_PATH):\n#         try:\n#             os.remove(DB_PATH)\n#         except Exception as e:\n#             st.error(f\"Could not remove DB file: {e}\")\n#         # Recreate table with required columns so UI continues to work\n#         ensure_table_and_columns()\n#         st.success(\"Database cleared.\")\n#     else:\n#         st.info(\"Database already empty.\")\n\n# # Show DB table\n# if st.checkbox(\"Show database records\"):\n#     # Ensure schema present and then show up to 100 rows\n#     ensure_table_and_columns()\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions ORDER BY id DESC LIMIT 100\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         df = pd.DataFrame()\n#     conn.close()\n#     if df.empty:\n#         st.info(\"No records found.\")\n#     else:\n#         st.dataframe(df)\n"}, "1d0c1d3868a9d2b3a27a87820ecde96a40d1da53": {"t": 1785753084.4074812, "n": 13174, "h": "53da1fdb056298b3", "c": "# # app.py (merged, schema-migrating version — fixes \"no such column\" errors)\n# import streamlit as st\n# from PIL import Image\n# from img_classification import (\n#     teachable_machine_classification,\n#     grad_cam_for_pil,\n#     is_image_blurry,\n#     prepare_result_record,\n#     insert_record,\n#     get_model_metadata,\n# )\n# import base64\n# import os\n# import requests\n# from dotenv import load_dotenv\n# from io import BytesIO\n# import json\n# import matplotlib.pyplot as plt\n# import tempfile\n# import sqlite3\n# import pandas as pd\n\n# load_dotenv()\n\n# DEEPSEEK_API_KEY = os.getenv(\"DEEPSEEK_API_KEY\")\n# DEEPSEEK_MODEL = os.getenv(\"DEEPSEEK_MODEL\", \"deepseek-chat\")\n# DEEPSEEK_BASE = os.getenv(\"DEEPSEEK_BASE_URL\", \"https://api.deepseek.com/v1\")\n\n# st.set_page_config(page_title=\"Breast Cancer Classifier\", layout=\"wide\")\n\n# DB_PATH = os.path.join(os.path.dirname(__file__), \"predictions.db\")\n\n# # --------------------------------------------------------------------\n# # DATABASE MIGRATION / ENSURE SCHEMA\n# # --------------------------------------------------------------------\n# REQUIRED_COLUMNS = {\n#     \"id\": \"INTEGER PRIMARY KEY AUTOINCREMENT\",\n#     \"timestamp\": \"TEXT\",\n#     \"patient_name\": \"TEXT\",\n#     \"patient_age\": \"INTEGER\",\n#     \"patient_gender\": \"TEXT\",\n#     \"notes\": \"TEXT\",\n#     \"image_paths\": \"TEXT\",\n#     \"prediction\": \"TEXT\",\n#     \"probs\": \"TEXT\",\n#     \"explanation\": \"TEXT\",\n#     \"model_metadata\": \"TEXT\",\n#     \"psno\": \"TEXT\"\n# }\n\n# def ensure_table_and_columns():\n#     \"\"\"\n#     Ensures `predictions` table exists and contains all REQUIRED_COLUMNS.\n#     If table exists but misses columns, add them via ALTER TABLE.\n#     \"\"\"\n#     conn = sqlite3.connect(DB_PATH)\n#     c = conn.cursor()\n\n#     # If table doesn't exist, create with full schema\n#     c.execute(\"SELECT name FROM sqlite_master WHERE type='table' AND name='predictions'\")\n#     if not c.fetchone():\n#         # build create statement using REQUIRED_COLUMNS in a deterministic order\n#         cols_sql = \",\\n    \".join([f\"{col} {REQUIRED_COLUMNS[col]}\" for col in REQUIRED_COLUMNS])\n#         create_sql = f\"CREATE TABLE predictions (\\n    {cols_sql}\\n)\"\n#         c.execute(create_sql)\n#         conn.commit()\n#         conn.close()\n#         return\n\n#     # Table exists: inspect columns\n#     c.execute(\"PRAGMA table_info(predictions)\")\n#     existing = [row[1] for row in c.fetchall()]  # row format: (cid, name, type, notnull, dflt_value, pk)\n#     # Add any missing columns\n#     for col, coldef in REQUIRED_COLUMNS.items():\n#         if col not in existing:\n#             try:\n#                 # For PK we skip altering since it's only for creation; but if id absent, we still add a column id (non-PK)\n#                 if col == \"id\":\n#                     # If id missing, add a plain INTEGER column (can't make it PK on alter)\n#                     c.execute(f\"ALTER TABLE predictions ADD COLUMN {col} INTEGER\")\n#                 else:\n#                     c.execute(f\"ALTER TABLE predictions ADD COLUMN {col} {coldef}\")\n#             except Exception:\n#                 # best-effort: ignore failures (e.g., column already added by race condition)\n#                 pass\n\n#     conn.commit()\n#     conn.close()\n\n# # ensure schema before anything\n# ensure_table_and_columns()\n\n# # --------------------------------------------------------------------\n# # SIDEBAR PATIENT INFO (includes PSNO)\n# # --------------------------------------------------------------------\n# st.sidebar.header(\"Patient Details\")\n# with st.sidebar.form(\"patient_form\"):\n#     psno = st.text_input(\"Patient PSNO (unique ID)\")\n#     name = st.text_input(\"Patient name\")\n#     age = st.number_input(\"Age\", min_value=0, max_value=130, value=40)\n#     gender = st.selectbox(\"Gender\", [\"Female\", \"Male\", \"Other\"])\n#     symptoms = st.text_area(\"Clinical notes (symptoms)\")\n#     submit_patient = st.form_submit_button(\"Save patient info\")\n\n# if submit_patient:\n#     if not psno:\n#         st.sidebar.error(\"PSNO is required.\")\n#     else:\n#         st.sidebar.success(\"Patient info saved (temporary)\")\n\n# # --------------------------------------------------------------------\n# # MODEL METADATA (unchanged)\n# # --------------------------------------------------------------------\n# st.sidebar.header(\"Model Info\")\n# meta = get_model_metadata()\n# st.sidebar.write(f\"Model file: {os.path.basename(meta['model_path'])}\")\n# st.sidebar.write(f\"Input size: {meta['input_size']}\")\n# st.sidebar.write(f\"Labels: {meta['labels']}\")\n# st.sidebar.write(f\"Loaded layers: {len(meta['keras_model_layers'])} layers\")\n\n# # --------------------------------------------------------------------\n# # MAIN: Upload, classify, grad-cam, LLM, insert\n# # --------------------------------------------------------------------\n# st.title(\"🩺 Breast Cancer Ultrasound Classifier — Enhanced\")\n# st.markdown(\"Upload ultrasound images for automated analysis, Grad-CAM, explanation, and DB saving.\")\n\n# uploaded_files = st.file_uploader(\"Upload image(s)\", accept_multiple_files=True, type=[\"jpg\",\"jpeg\",\"png\"])\n# exp_mode = st.radio(\"Explanation mode\", (\"Doctor\", \"Patient (plain language)\"))\n\n# if uploaded_files:\n#     results = []\n#     temp_dir = tempfile.mkdtemp(prefix=\"bcapp2_\")\n\n#     for uploaded in uploaded_files:\n#         try:\n#             image = Image.open(uploaded).convert(\"RGB\")\n#         except Exception:\n#             st.error(f\"Could not open {uploaded.name}\")\n#             continue\n\n#         img_path = os.path.join(temp_dir, uploaded.name)\n#         image.save(img_path)\n\n#         blurry, lap_var = is_image_blurry(image)\n#         if blurry:\n#             st.warning(f\"{uploaded.name} appears blurry (Laplacian variance={lap_var:.1f}).\")\n\n#         pred_idx, probs = teachable_machine_classification(image)\n#         class_names = [\"Malignant\", \"Benign\"]\n#         pred_label = class_names[pred_idx]\n\n#         # GRAD-CAM\n#         try:\n#             overlay, heatmap = grad_cam_for_pil(image)\n#             grad_path = os.path.join(temp_dir, f\"gradcam_{uploaded.name}\")\n#             overlay.save(grad_path)\n#         except Exception:\n#             grad_path = None\n\n#         # LLM (keep behavior from app.py2)\n#         explanation = \"LLM disabled (no API key configured).\"\n#         if DEEPSEEK_API_KEY:\n#             headers = {\"Authorization\": f\"Bearer {DEEPSEEK_API_KEY}\", \"Content-Type\": \"application/json\"}\n#             payload = {\n#                 \"model\": DEEPSEEK_MODEL,\n#                 \"messages\": [\n#                     {\"role\": \"system\", \"content\": \"You are a medical assistant.\"},\n#                     {\"role\": \"user\", \"content\": (\n#                         f\"A breast ultrasound image was analyzed by a machine learning model.\\n\"\n#                         f\"Prediction: {pred_label}\\n\"\n#                         f\"Probabilities: Malignant {probs[0]*100:.2f}%, Benign {probs[1]*100:.2f}%.\\n\"\n#                         f\"Patient name: {name}\\nAge: {age}\\nGender: {gender}\\nNotes: {symptoms}\\n\"\n#                         f\"Mode: {exp_mode}\"\n#                     )}\n#                 ],\n#                 \"max_tokens\": 700,\n#                 \"temperature\": 0.2\n#             }\n#             try:\n#                 r = requests.post(f\"{DEEPSEEK_BASE}/chat/completions\", headers=headers, json=payload, timeout=30)\n#                 if r.status_code == 200:\n#                     data = r.json()\n#                     explanation = data[\"choices\"][0][\"message\"][\"content\"].strip()\n#                 else:\n#                     explanation = f\"DeepSeek error: {r.status_code} {r.text}\"\n#             except Exception as e:\n#                 explanation = f\"DeepSeek API call failed: {e}\"\n\n#         # Recommendation logic (same)\n#         recommendation = \"\"\n#         if probs[0] >= 0.70:\n#             recommendation = \"High suspicion for malignancy — urgent follow-up required.\"\n#         elif probs[1] >= 0.90:\n#             recommendation = \"High confidence benign — routine follow-up.\"\n#         elif max(probs) < 0.55:\n#             recommendation = \"Low confidence — further imaging advised.\"\n\n#         # Prepare record using original prepare_result_record (keeps compatibility)\n#         patient_info = {\"name\": name, \"age\": age, \"gender\": gender, \"notes\": symptoms}\n#         try:\n#             rec = prepare_result_record(patient_info, [img_path], pred_label, probs, explanation)\n#             # rec has fields: timestamp, patient_name, patient_age, patient_gender, notes, image_paths,\n#             #                 prediction, probs, explanation, model_metadata\n#             # Add psno into record so insert_record will store it in the DB's psno column (if insert_record supports it)\n#             rec[\"psno\"] = psno\n#             # Use the original insert_record() from img_classification.py\n#             insert_record(rec)\n#         except Exception as e:\n#             st.error(f\"DB insert error: {e}\")\n\n#         # collect for display\n#         results.append({\n#             \"filename\": uploaded.name,\n#             \"image_path\": img_path,\n#             \"gradcam_path\": grad_path,\n#             \"prediction\": pred_label,\n#             \"probs\": [float(probs[0]), float(probs[1])],\n#             \"explanation\": explanation,\n#             \"recommendation\": recommendation\n#         })\n\n#     # Display results\n#     st.markdown(\"### Results\")\n#     for r in results:\n#         st.markdown(f\"**{r['filename']}** — Prediction: **{r['prediction']}**\")\n#         st.write(f\"Malignant: {r['probs'][0]*100:.2f}% | Benign: {r['probs'][1]*100:.2f}%\")\n#         col1, col2 = st.columns(2)\n#         with col1:\n#             st.image(r[\"image_path\"], caption=\"Original\")\n#         with col2:\n#             if r[\"gradcam_path\"]:\n#                 st.image(r[\"gradcam_path\"], caption=\"Grad-CAM\")\n#         st.info(r[\"recommendation\"])\n#         with st.expander(\"Explanation\"):\n#             st.write(r[\"explanation\"])\n#         fig, ax = plt.subplots(figsize=(4,1.2))\n#         ax.barh([\"Malignant\",\"Benign\"], [r['probs'][0]*100, r['probs'][1]*100])\n#         ax.set_xlim(0,100)\n#         st.pyplot(fig)\n#         st.markdown(\"---\")\n\n# # --------------------------------------------------------------------\n# # DATABASE MANAGEMENT UI (replaces PDF export)\n# # --------------------------------------------------------------------\n# st.header(\"📦 Database Management\")\n\n# # Download CSV\n# if st.button(\"Download database as CSV\"):\n#     ensure_table_and_columns = None  # placeholder to avoid linter confusion (we used ensure_table_and_columns earlier)\n#     # ensure schema and read\n#     ensure_table_and_columns = None\n#     # Re-run the ensure (safe to call)\n#     ensure_table_and_columns = lambda: ensure_table_and_columns  # no-op\n#     # We'll just make sure the DB exists and then read\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         conn.close()\n#         df = pd.DataFrame()\n#     conn.close()\n\n#     if df.empty:\n#         st.info(\"Database is empty.\")\n#     else:\n#         csv = df.to_csv(index=False).encode()\n#         b64 = base64.b64encode(csv).decode()\n#         st.markdown(f'<a href=\"data:file/csv;base64,{b64}\" download=\"predictions.csv\">Download CSV</a>', unsafe_allow_html=True)\n\n# # Download Excel\n# if st.button(\"Download database as Excel\"):\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         conn.close()\n#         df = pd.DataFrame()\n#     conn.close()\n\n#     if df.empty:\n#         st.info(\"Database is empty.\")\n#     else:\n#         excel_path = os.path.join(tempfile.gettempdir(), \"predictions.xlsx\")\n#         df.to_excel(excel_path, index=False)\n#         with open(excel_path, \"rb\") as f:\n#             excel_bytes = f.read()\n#             b64 = base64.b64encode(excel_bytes).decode()\n#             st.markdown(f'<a href=\"data:application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;base64,{b64}\" download=\"predictions.xlsx\">Download Excel</a>', unsafe_allow_html=True)\n\n# # Clear DB\n# if st.button(\"Clear entire database\"):\n#     if os.path.exists(DB_PATH):\n#         try:\n#             os.remove(DB_PATH)\n#         except Exception as e:\n#             st.error(f\"Could not remove DB file: {e}\")\n#         # Recreate table with required columns so UI continues to work\n#         ensure_table_and_columns()\n#         st.success(\"Database cleared.\")\n#     else:\n#         st.info(\"Database already empty.\")\n\n# # Show DB table\n# if st.checkbox(\"Show database records\"):\n#     # Ensure schema present and then show up to 100 rows\n#     ensure_table_and_columns()\n#     conn = sqlite3.connect(DB_PATH)\n#     try:\n#         df = pd.read_sql_query(\"SELECT * FROM predictions ORDER BY id DESC LIMIT 100\", conn)\n#     except Exception as e:\n#         st.error(f\"Failed to read DB: {e}\")\n#         df = pd.DataFrame()\n#     conn.close()\n#     if df.empty:\n#         st.info(\"No records found.\")\n#     else:\n#         st.dataframe(df)\n"}}}}