app.py

import streamlit as st
import torch
import torch.nn as nn
import math

# Set page config at the very beginning
st.set_page_config(page_title='Total Corneal Astigmatism Prediction', page_icon=':eyeglasses:', layout='wide')

# Custom CSS to set background color to #000 for main app and navigation
st.markdown("""
<style>
    .stApp {
        background-color: #000;
    }
    
    .stDeployButton {
        display: none !important;
    }

    header[data-testid="stHeader"] {
        background-color: #000;
    }

    .stDecoration {
        background-color: #000 !important;
    }
    
    .stToolbar {
        background-color: #000 !important;
    }

    #MainMenu {
        background-color: #000 !important;
    }

    div[data-testid="stToolbar"] {
        background-color: #000 !important;
    }

    button[kind="headerNoPadding"], button[data-testid="baseButton-headerNoPadding"], button[aria-haspopup="menu"] {
        background-color: transparent !important;
    }

    .stApp > header {
        background-color: transparent !important;
    }




    .stTextInput > div > div > input {
        background-color: #333 !important;
        color: white !important;
    }
    .stNumberInput > div > div > input {
        background-color: #333 !important;
        color: white !important;
    }
    .stTextInput > label, .stNumberInput > label {
        color: white !important;
        font-size: 24px !important;
    }
    [data-testid="stNumberInput"] {
        margin-top: -15px;
    }
</style>
""", unsafe_allow_html=True)

class RegressionModel2(nn.Module):
    def __init__(self, input_dim2, hidden_dim2, output_dim2):
        super(RegressionModel2, self).__init__()
        self.fc1 = nn.Linear(input_dim2, hidden_dim2)
        self.relu1 = nn.ReLU()
        self.fc2 = nn.Linear(hidden_dim2, output_dim2)
        self.batch_norm1 = nn.BatchNorm1d(hidden_dim2)

    def forward(self, x2):
        out = self.fc1(x2)
        out = self.relu1(out)
        out = self.batch_norm1(out)
        out = self.fc2(out)
        return out

@st.cache_resource
def load_models():
    model_j0 = RegressionModel2(3, 32, 1)
    model_j0.load_state_dict(torch.load('j0_model-2.pt'))
    model_j0.eval()

    model_j45 = RegressionModel2(3, 32, 1)
    model_j45.load_state_dict(torch.load('j45_model-2.pt'))
    model_j45.eval()
    
    return model_j0, model_j45

model_j0, model_j45 = load_models()

def calculate_initial_j0_j45(magnitude, axis_deg):
    axis_rad = math.radians(axis_deg)
    j0 = magnitude * math.cos(2 * axis_rad)
    j45 = magnitude * math.sin(2 * axis_rad)
    return j0, j45

def predict_new_j0_j45(age, aca_magnitude, aca_axis_deg):
    initial_j0, initial_j45 = calculate_initial_j0_j45(aca_magnitude, aca_axis_deg)
    
    input_data_j0 = torch.tensor([[age, aca_axis_deg, initial_j0]], dtype=torch.float32)
    input_data_j45 = torch.tensor([[age, aca_axis_deg, initial_j45]], dtype=torch.float32)
    
    with torch.no_grad():
        new_j0 = model_j0(input_data_j0).item()
        new_j45 = model_j45(input_data_j45).item()
    
    return new_j0, new_j45

def main():
    #st.title('Total Corneal Astigmatism Prediction')

    # Initialize session state for input values if not already present
    if 'age' not in st.session_state:
        st.session_state.age = None
    if 'aca_magnitude' not in st.session_state:
        st.session_state.aca_magnitude = None
    if 'aca_axis' not in st.session_state:
        st.session_state.aca_axis = None

    # Input fields using session state
    st.markdown('<p style="font-size: 20px; color: white; margin-bottom: 0px;">Enter Patient Age (18-90 Years):</p>', unsafe_allow_html=True)
    #age = st.number_input('Enter Patient Age (18-90 Years):',
    age = st.number_input('',
                          min_value=18.0, max_value=90.0, 
                          value=st.session_state.age if st.session_state.age is not None else None,
                          step=0.1, 
                          key='age')
    st.markdown('<p style="font-size: 20px; color: white; margin-bottom: 0px;">Enter ACA Magnitude (0-10 Diopters):</p>', unsafe_allow_html=True)
    #aca_magnitude = st.number_input('Enter ACA Magnitude (0-10 Diopters):',
    aca_magnitude = st.number_input('', 
                                    min_value=0.0, max_value=10.0, 
                                    value=st.session_state.aca_magnitude if st.session_state.aca_magnitude is not None else None,
                                    step=0.01, 
                                    key='aca_magnitude')
    st.markdown('<p style="font-size: 20px; color: white; margin-bottom: 0px;">Enter ACA Axis (0-180 Degrees):</p>', unsafe_allow_html=True)
    #aca_axis = st.number_input('Enter ACA Axis (0-180 Degrees):',
    aca_axis = st.number_input('', 
                               min_value=0.0, max_value=180.0, 
                               value=st.session_state.aca_axis if st.session_state.aca_axis is not None else None,
                               step=0.1, 
                               key='aca_axis')

    if st.button('Predict!'):
        if age is not None and aca_magnitude is not None and aca_axis is not None:
            if 18 <= age <= 90 and 0 <= aca_magnitude <= 10 and 0 <= aca_axis <= 180:
                # Calculate initial J0 and J45
                initial_j0, initial_j45 = calculate_initial_j0_j45(aca_magnitude, aca_axis)
                
                # Predict new J0 and J45 using the models
                new_j0, new_j45 = predict_new_j0_j45(age, aca_magnitude, aca_axis)
                
                # Calculate predicted magnitude and axis
                predicted_magnitude = math.sqrt(new_j0**2 + new_j45**2)
                predicted_axis = 0.5 * math.degrees(math.atan2(new_j45, new_j0))
                if predicted_axis < 0:
                    predicted_axis += 180
                
                # Display results in green success boxes
                st.success(f'Predicted Total Corneal Astigmatism Magnitude: {predicted_magnitude:.2f} D')
                st.success(f'Predicted Total Corneal Astigmatism Axis: {predicted_axis:.1f}°')
                
                # Display intermediate values for verification
                #st.info(f'''
                #Input ACA - Magnitude: {aca_magnitude:.2f} D, Axis: {aca_axis:.1f}°
                #Initial J0: {initial_j0:.2f}, Initial J45: {initial_j45:.2f}
                #Predicted J0: {new_j0:.2f}, Predicted J45: {new_j45:.2f}
                #''')
                
                # Additional debugging information (optional)
                #if st.checkbox('Show detailed model inputs and outputs'):
                    #st.subheader('Debugging Information:')
                    #st.write(f"Input tensor for J0: {torch.tensor([[age, aca_axis, initial_j0]], dtype=torch.float32)}")
                    #st.write(f"Input tensor for J45: {torch.tensor([[age, aca_axis, initial_j45]], dtype=torch.float32)}")
                    #st.write(f"Raw J0 output: {new_j0}")
                    #st.write(f"Raw J45 output: {new_j45}")
            else:
                st.error('Please ensure all inputs are within the specified ranges.')
        else:
            st.error('Please fill in all input fields before predicting.')

if __name__ == '__main__':
    main()