Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,156 +1,238 @@
+"""
+🏠 ICD-10 Automated Clinical Coding System — Dashboard
+"""
 import streamlit as st
-import pandas as pd
-import numpy as np
-import joblib
-from scipy.special import softmax
+import os
+
+# ── Page Config ──
+st.set_page_config(
+    page_title="ICD-10 AutoCoder",
+    page_icon="🏥",
+    layout="wide",
+    initial_sidebar_state="expanded",
+)
+
+# ── Load CSS ──
+css_path = os.path.join(os.path.dirname(__file__), "assets", "style.css")
+if os.path.exists(css_path):
+    with open(css_path, encoding="utf-8") as f:
+        st.markdown(f"<style>{f.read()}</style>", unsafe_allow_html=True)
+
+from utils.config import DATASET_STATS, TOP_20_CODES
+
+# ── Sidebar ──
+with st.sidebar:
+    st.markdown("## 🏥 ICD-10 AutoCoder")
+    st.markdown("---")
+    st.markdown("""
+    **Navigation**
+    - 🏠 Dashboard (this page)
+    - 🔬 Live ICD Coding
+    - 📊 Model Comparison
+    - 📖 ICD-10 Browser
+    - 📈 Dataset Explorer
+    - 📂 Batch Prediction
+    - ⚙️ About
+    """)
+    st.markdown("---")
+    st.markdown(
+        '<p style="color: #64748b; font-size: 0.75rem;">'
+        'Built for clinical NLP research<br>'
+        'RTX 4050 · 6GB VRAM · LoRA</p>',
+        unsafe_allow_html=True
+    )
 
-# ── Load Models ─────────────────────────────────────
-@st.cache_resource
-def load_models():
-    svm = joblib.load("tfidf_svm_pipeline.joblib")
-    label_map = joblib.load("label_to_code.joblib")
 
-    # Import torch-based models safely
-    try:
-        from utils.models import predict_b2, predict_longformer
-        TORCH_OK = True
-    except Exception as e:
-        TORCH_OK = False
-        predict_b2 = None
-        predict_longformer = None
-
-    return svm, label_map, predict_b2, predict_longformer, TORCH_OK
-
-
-svm, label_map, predict_b2, predict_longformer, TORCH_OK = load_models()
+# ═══════════════════════════════════════════════════
+# HERO SECTION
+# ═══════════════════════════════════════════════════
+st.markdown("""
+<div class="hero-container">
+    <div class="hero-title">ICD-10 Automated Clinical Coding</div>
+    <div class="hero-subtitle">
+        A hierarchical retrieval and biomedical re-ranking system for automated 
+        ICD-10 diagnosis code assignment from clinical discharge notes. 
+        Bilingual output (English + Traditional Chinese) across 2,863 codes.
+    </div>
+</div>
+""", unsafe_allow_html=True)
 
-# ── SVM Prediction ───────────────────────────────────
-def predict_svm(text):
-    scores = svm.decision_function([text])[0]
-    probs = softmax(scores)
-    pred = int(np.argmax(probs))
-    return pred, probs
 
+# ═══════════════════════════════════════════════════
+# STAT CARDS
+# ═══════════════════════════════════════════════════
+st.markdown('<div class="section-header"><div class="section-accent"></div>Project at a Glance</div>',
+            unsafe_allow_html=True)
+
+c1, c2, c3, c4 = st.columns(4)
+
+with c1:
+    st.markdown(f"""
+    <div class="stat-card">
+        <div class="stat-number">{DATASET_STATS['raw_records']:,}</div>
+        <div class="stat-label">Clinical Records</div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with c2:
+    st.markdown(f"""
+    <div class="stat-card">
+        <div class="stat-number">{DATASET_STATS['unique_encounters']:,}</div>
+        <div class="stat-label">Patient Encounters</div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with c3:
+    st.markdown(f"""
+    <div class="stat-card">
+        <div class="stat-number">{DATASET_STATS['unique_codes']:,}</div>
+        <div class="stat-label">ICD-10 Codes</div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with c4:
+    st.markdown("""
+    <div class="stat-card">
+        <div class="stat-number">4</div>
+        <div class="stat-label">Trained Models</div>
+    </div>
+    """, unsafe_allow_html=True)
 
-# ── UI ──────────────────────────────────────────────
-st.title("🏥 ICD-10 Model Comparison System")
 
 # ═══════════════════════════════════════════════════
-# 🔬 SINGLE INPUT
+# ARCHITECTURE PIPELINE
 # ═══════════════════════════════════════════════════
-st.markdown("## 🔬 Single Clinical Note")
-
-text_input = st.text_area("Enter clinical note")
-
-if st.button("Run Prediction"):
-
-    if not text_input.strip():
-        st.warning("Enter text first")
-    else:
-        # SVM
-        svm_pred, _ = predict_svm(text_input)
-        svm_code = label_map[svm_pred]
-
-        # BERT
-        if TORCH_OK:
-            try:
-                bert_pred, _ = predict_b2(text_input)
-                bert_code = label_map[bert_pred]
-            except:
-                bert_code = "ERROR"
-        else:
-            bert_code = "NOT AVAILABLE"
-
-        # Longformer
-        if TORCH_OK:
-            try:
-                long_pred, _ = predict_longformer(text_input)
-                long_code = label_map[long_pred]
-            except:
-                long_code = "ERROR"
-        else:
-            long_code = "NOT AVAILABLE"
-
-        # Display
-        st.markdown("### 🎯 Predictions")
-        st.write(f"SVM → {svm_code}")
-        st.write(f"ClinicalBERT → {bert_code}")
-        st.write(f"Longformer → {long_code}")
-
-        # AGREEMENT
-        codes = [svm_code, bert_code, long_code]
-        valid_codes = [c for c in codes if c not in ["ERROR", "NOT AVAILABLE"]]
-
-        if len(set(valid_codes)) == 1:
-            st.success("✅ All models agree")
-        elif len(set(valid_codes)) == 2:
-            st.warning("⚠️ Partial agreement")
-        else:
-            st.error("❌ No agreement")
+st.markdown("")
+st.markdown('<div class="section-header"><div class="section-accent"></div>Three-Layer Pipeline Architecture</div>',
+            unsafe_allow_html=True)
+
+col_l, col_r = st.columns([3, 2])
+
+with col_l:
+    st.markdown("""
+    <div class="pipeline-step">
+        <span class="badge badge-teal">LAYER 1</span>
+        <div style="margin-top:0.5rem; font-size:1.05rem; font-weight:600; color:#e8ecf4;">
+            Hierarchical TF-IDF Retriever
+        </div>
+        <div style="font-size:0.85rem; color:#94a3b8; margin-top:0.3rem;">
+            Chapter → Category → Full code scoring<br>
+            Reduces 2,863 codes → top-100 candidates
+        </div>
+    </div>
+    <div class="pipeline-arrow">▼</div>
+    <div class="pipeline-step">
+        <span class="badge badge-blue">LAYER 2</span>
+        <div style="margin-top:0.5rem; font-size:1.05rem; font-weight:600; color:#e8ecf4;">
+            ClinicalBERT Re-Ranker (LoRA)
+        </div>
+        <div style="font-size:0.85rem; color:#94a3b8; margin-top:0.3rem;">
+            Pairwise relevance scoring: (note, ICD description) → score<br>
+            92.5% pairwise F1 accuracy
+        </div>
+    </div>
+    <div class="pipeline-arrow">▼</div>
+    <div class="pipeline-step">
+        <span class="badge badge-purple">LAYER 3</span>
+        <div style="margin-top:0.5rem; font-size:1.05rem; font-weight:600; color:#e8ecf4;">
+            Bilingual Prediction Output
+        </div>
+        <div style="font-size:0.85rem; color:#94a3b8; margin-top:0.3rem;">
+            Ranked ICD codes with confidence scores<br>
+            English + Traditional Chinese descriptions
+        </div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with col_r:
+    st.markdown("""
+    <div class="model-card" style="margin-bottom:1rem;">
+        <div class="model-card-label">BEST BASELINE</div>
+        <div class="model-card-title">TF-IDF + LinearSVC</div>
+        <div class="model-card-metric">0.516</div>
+        <div class="model-card-label">Micro-F1 (Top-100)</div>
+    </div>
+    <div class="model-card" style="margin-bottom:1rem;">
+        <div class="model-card-label">BEST TRANSFORMER</div>
+        <div class="model-card-title">ClinicalBERT (LoRA)</div>
+        <div class="model-card-metric">0.397</div>
+        <div class="model-card-label">Micro-F1 (Full 2,863)</div>
+    </div>
+    <div class="model-card">
+        <div class="model-card-label">PROPOSED PIPELINE</div>
+        <div class="model-card-title">Hierarchical Re-Ranker</div>
+        <div class="model-card-metric">0.925</div>
+        <div class="model-card-label">Pairwise Re-ranking F1</div>
+    </div>
+    """, unsafe_allow_html=True)
 
 
 # ═══════════════════════════════════════════════════
-# 📂 CSV BATCH MODE
+# QUICK ACCESS
 # ═══════════════════════════════════════════════════
-st.markdown("## 📂 Batch Testing (CSV)")
-
-uploaded_file = st.file_uploader("Upload CSV", type=["csv"])
-
-if uploaded_file:
-    df = pd.read_csv(uploaded_file)
-
-    if "text" not in df.columns:
-        st.error("CSV must contain 'text' column")
-    else:
-        results = []
-
-        for text in df["text"].astype(str):
-
-            # SVM
-            svm_pred, _ = predict_svm(text)
-            svm_code = label_map[svm_pred]
-
-            # BERT
-            if TORCH_OK:
-                try:
-                    bert_pred, _ = predict_b2(text)
-                    bert_code = label_map[bert_pred]
-                except:
-                    bert_code = "ERR"
-            else:
-                bert_code = "NA"
-
-            # Longformer
-            if TORCH_OK:
-                try:
-                    long_pred, _ = predict_longformer(text)
-                    long_code = label_map[long_pred]
-                except:
-                    long_code = "ERR"
-            else:
-                long_code = "NA"
-
-            # Agreement
-            codes = [svm_code, bert_code, long_code]
-            valid = [c for c in codes if c not in ["ERR", "NA"]]
-            agree = len(set(valid)) == 1
-
-            results.append({
-                "text": text[:80],
-                "svm": svm_code,
-                "bert": bert_code,
-                "longformer": long_code,
-                "agree": agree
-            })
-
-        res_df = pd.DataFrame(results)
-
-        st.dataframe(res_df)
-
-        # Metrics
-        agreement_rate = res_df["agree"].mean() * 100
-        st.markdown(f"### 📊 Agreement Rate: {agreement_rate:.2f}%")
-
-        # Download
-        csv = res_df.to_csv(index=False).encode('utf-8')
-        st.download_button("Download Results", csv, "results.csv", "text/csv")
+st.markdown("")
+st.markdown('<div class="section-header"><div class="section-accent"></div>Quick Access</div>',
+            unsafe_allow_html=True)
+
+qa1, qa2, qa3, qa4 = st.columns(4)
+
+with qa1:
+    st.markdown("""
+    <div class="stat-card">
+        <div style="font-size:2rem;">🔬</div>
+        <div style="font-size:1.05rem; font-weight:600; color:#e8ecf4; margin-top:0.5rem;">
+            Live ICD Coding
+        </div>
+        <div style="font-size:0.8rem; color:#94a3b8; margin-top:0.3rem;">
+            Paste a clinical note and get instant ICD-10 predictions
+        </div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with qa2:
+    st.markdown("""
+    <div class="stat-card">
+        <div style="font-size:2rem;">📊</div>
+        <div style="font-size:1.05rem; font-weight:600; color:#e8ecf4; margin-top:0.5rem;">
+            Model Comparison
+        </div>
+        <div style="font-size:0.8rem; color:#94a3b8; margin-top:0.3rem;">
+            Compare all 4 models with interactive charts
+        </div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with qa3:
+    st.markdown("""
+    <div class="stat-card">
+        <div style="font-size:2rem;">📖</div>
+        <div style="font-size:1.05rem; font-weight:600; color:#e8ecf4; margin-top:0.5rem;">
+            ICD-10 Browser
+        </div>
+        <div style="font-size:0.8rem; color:#94a3b8; margin-top:0.3rem;">
+            Search and explore all 2,863 ICD-10 codes
+        </div>
+    </div>
+    """, unsafe_allow_html=True)
+
+with qa4:
+    st.markdown("""
+    <div class="stat-card">
+        <div style="font-size:2rem;">📂</div>
+        <div style="font-size:1.05rem; font-weight:600; color:#e8ecf4; margin-top:0.5rem;">
+            Batch Prediction
+        </div>
+        <div style="font-size:0.8rem; color:#94a3b8; margin-top:0.3rem;">
+            Upload CSV files for bulk ICD-10 coding with multi-model comparison
+        </div>
+    </div>
+    """, unsafe_allow_html=True)
+
+
+# ── Footer ──
+st.markdown("""
+<div class="footer">
+    ICD-10 Automated Coding System · Biomedical NLP Research · 2026<br>
+    Built with ClinicalBERT · Longformer · LoRA · PyTorch · Streamlit
+</div>
+""", unsafe_allow_html=True)