Update app.py

app.py

@@ -1012,74 +1012,119 @@ FIGSIZE = (plot_width, plot_height)
 st.title("Explainable-Acute-Leukemia-Mortality-Predictor")
 st.caption("Explainable clinical AI for mortality and outcome prediction in acute leukemia using SHAP-interpretable models")
 
-with st.expander("About this framework, who can use it, and required Excel format", expanded=True):
+with st.expander("About this AI model, who can use it, and required Excel format", expanded=True):
     st.markdown("""
-## What is this framework?
-This is a **no-code, explainable machine-learning MODEL** that allows you to upload an Excel sheet, train a predictive model instantly, and obtain **transparent, variable-level explanations** of the prediction.  
-
-You provide:
-- Predictor variables as Excel columns
-- A binary outcome in column **`Outcome Event`**
-
-The system automatically:
-- Trains a validated logistic regression–based model
-- Performs internal and external validation (if labels are present)
-- Generates **explainability (SHAP)** showing which variables contribute most to predictions
-- Allows reuse of the trained model on new Excel sheets with identical column names
-
----
-
-## Who can use this?
-This framework is designed for:
-- **Clinicians and physician-scientists**
-- **Clinical researchers**
-- **Epidemiologists and outcomes researchers**
-- **Health-data and AI researchers**
-
-No programming or machine-learning expertise is required.  
-All modeling, validation, and explainability are handled automatically.
-
----
-
-## Training Excel (with labels)
-- **First row must contain column names**
-- **All columns except `Outcome Event`** → model input features (predictors)
-- **`Outcome Event`** → binary outcome label to be predicted  
-  - Accepted formats: `0/1`, `Yes/No`, `True/False`
-
----
-
-## Variable type selection (during training)
-- You will explicitly choose which predictors are:
-  - **Numeric** (median imputation + scaling)
-  - **Categorical** (most-frequent imputation + one-hot encoding)
-- This variable-type schema is **saved with the trained model** and enforced during all future predictions.
-
----
-
-## Prediction / External validation Excel
-- Must contain the **same predictor columns** with **identical names** as the trained model
-- **Do NOT include `Outcome Event`** if you only want predictions
-- **Include `Outcome Event`** if you want **external validation metrics**, including:
-  - ROC AUC
-  - Sensitivity / specificity
-  - Precision–recall
-  - Calibration
-  - Decision curve analysis
-  - Confusion matrix
-
----
-
-## Explainability and downloads
-- For both training and validation, the framework provides **SHAP-based explanations** indicating:
-  - Which variables contribute most to each prediction
-  - Direction and magnitude of influence
-- You can download:
-  - **Prediction output sheets** (probabilities, classes, risk bands)
-  - **All plots individually** (ROC, PR, calibration, DCA, SHAP)
-  - Plots are exportable as **high-resolution PNG (≥600 DPI)** for publications
-
-""")
+    ## What is this framework?
+    
+    This is a **clinically oriented, explainable AI platform** for developing and validating **mortality and outcome prediction models in acute leukemia** using structured Excel data.
+    
+    The system integrates:
+    
+    • Statistical modeling (logistic regression)  
+    • Explainable AI (SHAP)  
+    • Bootstrap internal validation  
+    • External clinical validation  
+    • Publication-ready performance reporting  
+    
+    into a **single no-code workflow** designed for clinicians and researchers.
+    
+    The goal is to produce **transparent, trustworthy, and clinically interpretable predictions**, rather than black-box outputs.
+    
+    ---
+    
+    ## What does it do automatically?
+    
+    After uploading your Excel file, the platform will:
+    
+    ### Model development
+    • Train a logistic-regression–based clinical prediction model  
+    • Handle preprocessing automatically  
+     – numeric → imputation + scaling  
+     – categorical → imputation + one-hot encoding  
+    • Save the full schema to ensure reproducibility  
+
+    ### Validation
+    • ROC AUC and ROC curves  
+    • Precision–Recall curves  
+    • Calibration curves + Brier score  
+    • Decision Curve Analysis (clinical net benefit)  
+    • Sensitivity / specificity / F1 / balanced accuracy  
+    • Threshold optimisation  
+    
+    ### Internal validation (recommended)
+    • Bootstrap out-of-bag validation (multiple resamples)  
+    • 95% confidence intervals for metrics  
+    • Reduced optimism bias for small clinical datasets  
+    
+    ### Explainability
+    • SHAP feature importance  
+    • Patient-level waterfall plots  
+    • Global and local explanations  
+    
+    ### Deployment
+    • One-click publishing of trained models  
+    • Reuse the same model on future Excel sheets  
+    • Download predictions, plots, and reports  
+    
+    All plots are exportable as **high-resolution (≥600 DPI) publication-ready figures**.
+    
+    ---
+
+    ## Who can use this?
+    
+    This framework is intended for:
+    
+    • Hematology–Oncology clinicians  
+    • Clinical researchers  
+    • Epidemiologists  
+    • Outcomes researchers  
+    • Students learning explainable AI  
+    
+    No programming or machine-learning expertise is required.
+    
+    ---
+    
+    ## Required Excel format
+    
+    ### Training file (with labels)
+    • First row must contain column names  
+    • All columns except **Outcome Event** → predictor variables  
+    • **Outcome Event** → binary label  
+     Accepted formats: 0/1, Yes/No, True/False  
+    
+    ### Variable type selection
+    During training you explicitly choose:
+    • Numeric variables  
+    • Categorical variables  
+    
+    This schema is saved with the model and **must match future files exactly**.
+    
+    ---
+
+    ### Prediction / External validation file
+    Must contain:
+    • Same predictor column names as the trained model  
+    
+    Optional:
+    • Include **Outcome Event** to compute full external validation metrics  
+    
+    If labels are included, the system will automatically generate:
+    • ROC  
+    • Calibration  
+    • Decision curves  
+    • Confusion matrix  
+    • Clinical performance metrics  
+    
+    ---
+    
+    ## Important note
+    
+    This tool is for **research and decision-support only**.  
+    It is **not a medical device** and must not replace clinical judgment.
+    """)
+    
+
+
 
 st.warning(
     "Prediction will fail if feature names or variable types "