app.py

import streamlit as st
import numpy as np
import onnxruntime as rt
import soundfile as sf
import sounddevice as sd
from scipy.io.wavfile import write

# Load the ONNX model
model_path = 'model.onnx'  # Replace with the actual path to your model
model_inference = rt.InferenceSession(model_path)

# Function to preprocess audio data
def preprocess_audio(audio_data):
    mu = np.nanmean(audio_data)
    std = np.nanstd(audio_data)
    audio_data = (audio_data - mu) / std
    audio_data = np.pad(audio_data, (0, 22050 - len(audio_data)), 'constant').reshape(1, -1, 1).astype(np.float32)
    return audio_data

# Function to preprocess clinical data
def preprocess_clinical_data(age, height, weight, reported_cough_dur, heart_rate, temperature, sex, tb_prior, tb_prior_Pul, tb_prior_Extrapul, tb_prior_Unknown, hemoptysis, weight_loss, smoke_lweek, fever, night_sweats):
    sex_Female = 1 if sex == 'Female' else 0
    sex_Male = 1 if sex == 'Male' else 0
    tb_prior_No = 1 if tb_prior == 'No' else 0
    tb_prior_Not_sure = 1 if tb_prior == 'Not sure' else 0
    tb_prior_Yes = 1 if tb_prior == 'Yes' else 0
    tb_prior_Pul_No = 1 if tb_prior_Pul == 'No' else 0
    tb_prior_Pul_Yes = 1 if tb_prior_Pul == 'Yes' else 0
    tb_prior_Extrapul_No = 1 if tb_prior_Extrapul == 'No' else 0
    tb_prior_Extrapul_Yes = 1 if tb_prior_Extrapul == 'Yes' else 0
    tb_prior_Unknown_No = 1 if tb_prior_Unknown == 'No' else 0
    tb_prior_Unknown_Yes = 1 if tb_prior_Unknown == 'Yes' else 0
    hemoptysis_No = 1 if hemoptysis == 'No' else 0
    hemoptysis_Yes = 1 if hemoptysis == 'Yes' else 0
    weight_loss_No = 1 if weight_loss == 'No' else 0
    weight_loss_Yes = 1 if weight_loss == 'Yes' else 0
    smoke_lweek_No = 1 if smoke_lweek == 'No' else 0
    smoke_lweek_Yes = 1 if smoke_lweek == 'Yes' else 0
    fever_No = 1 if fever == 'No' else 0
    fever_Yes = 1 if fever == 'Yes' else 0
    night_sweats_No = 1 if night_sweats == 'No' else 0
    night_sweats_Yes = 1 if night_sweats == 'Yes' else 0

    clinical_data = [age, height, weight, reported_cough_dur, heart_rate, temperature,
                     sex_Female, sex_Male, tb_prior_No, tb_prior_Not_sure, tb_prior_Yes,
                     tb_prior_Pul_No, tb_prior_Pul_Yes, tb_prior_Extrapul_No, tb_prior_Extrapul_Yes,
                     tb_prior_Unknown_No, tb_prior_Unknown_Yes, hemoptysis_No, hemoptysis_Yes,
                     weight_loss_No, weight_loss_Yes, smoke_lweek_No, smoke_lweek_Yes,
                     fever_No, fever_Yes, night_sweats_No, night_sweats_Yes]
    clinical_data = np.array(clinical_data).reshape(1, -1).astype(np.float32)
    return clinical_data

# Main function to run the app
def main():
    st.title('TB Cough Sound Analysis')

    # Create tabs
    tabs = ["Record Cough Sound", "Upload Cough Sound"]
    choice = st.sidebar.selectbox("Choose Option", tabs)

    if choice == "Record Cough Sound":
        st.write("**Record Cough Sound**")
        st.write("Press the button below to start recording:")
        start_recording = st.button("Start Recording")

        if start_recording:
            duration = 5  # Set the default recording duration to 5 seconds
            st.write("Recording started...")
            sample_rate = 22050
            audio_data = sd.rec(int(duration * sample_rate), samplerate=sample_rate, channels=1)
            sd.wait()
            write("audio_file.wav", sample_rate, audio_data)
            st.success("Recording saved as 'audio_file.wav'.")
            st.info("Please proceed to the next step.")

    elif choice == "Upload Cough Sound":
        st.write("**Upload Cough Sound**")
        uploaded_file = st.file_uploader("Upload Cough Sound (WAV file)", type=["wav"])
        if uploaded_file is not None:
            with open("audio_file.wav", "wb") as f:
                f.write(uploaded_file.getvalue())
            st.success("File uploaded successfully.")
            st.info("Please proceed to the next step.")

    st.write('**Step 2: Enter Clinical Information**')
    age = st.slider('Age', 1, 100, 30)
    height = st.slider('Height (cm)', 100, 300, 170)
    weight = st.slider('Weight (kg)', 20, 200, 70)
    reported_cough_dur = st.slider('Reported Cough Duration (days)', 1, 100, 10)
    heart_rate = st.slider('Heart Rate (bpm)', 50, 200, 80)
    temperature = st.slider('Body Temperature (°C)', 35.0, 40.0, 37.0)
    sex = st.radio('Sex', ('Male', 'Female'))
    tb_prior = st.radio('TB Prior', ('No', 'Not sure', 'Yes'))
    tb_prior_Pul = st.radio('TB Prior Pul', ('No', 'Yes'))
    tb_prior_Extrapul = st.radio('TB Prior Extrapul', ('No', 'Yes'))
    tb_prior_Unknown = st.radio('TB Prior Unknown', ('No', 'Yes'))
    hemoptysis = st.radio('Hemoptysis', ('No', 'Yes'))
    weight_loss = st.radio('Weight Loss', ('No', 'Yes'))
    smoke_lweek = st.radio('Smoke Lweek', ('No', 'Yes'))
    fever = st.radio('Fever', ('No', 'Yes'))
    night_sweats = st.radio('Night Sweats', ('No', 'Yes'))

    if st.button('Predict'):
        if choice == "Record Cough Sound":
            audio_file_path = "audio_file.wav"
        elif choice == "Upload Cough Sound":
            audio_file_path = "audio_file.wav"

        raw_values, rate = sf.read(audio_file_path)
        audio_data = preprocess_audio(raw_values)
        clinical_data = preprocess_clinical_data(age, height, weight, reported_cough_dur, heart_rate, temperature,
                                                 sex, tb_prior, tb_prior_Pul, tb_prior_Extrapul, tb_prior_Unknown,
                                                 hemoptysis, weight_loss, smoke_lweek, fever, night_sweats)
        input_name = model_inference.get_inputs()[0].name
        input_name2 = model_inference.get_inputs()[1].name
        label_name = model_inference.get_outputs()[0].name
        onnx_pred = model_inference.run([label_name], {input_name: audio_data, input_name2: clinical_data})
        result = onnx_pred[0]

        st.write(f"**Prediction:** {result[0][0]}")
        if result[0][0] >= 0.5:
            st.write("Tuberculosis (TB) is found based on the audio data and clinical information.")
        else:
            st.write("No Tuberculosis (TB) is found based on the audio data and clinical information.")

if __name__ == "__main__":
    main()