app.py

import streamlit as st
from utils import get_data, diseases, metrics_table
from transformers import pipeline
import pandas as pd 
import io
import json
from openai import OpenAI
import tensorflow as tf
from transformers import TFAutoModelForSequenceClassification, AutoTokenizer


# Load model
model_name = "Zabihin/Symptom_to_Diagnosis"
model = TFAutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)


def get_data(file:str):
    """
    Reads data from a JSON file and returns the loaded data.

    Parameters:
        file (str): The path to the JSON file to be read.

    Returns:
        dict: The data loaded from the specified JSON file.
    """
    with open(file) as files:
        data = json.load(files)
    return data


def call_gpt3(api,prompt:str):
    client = OpenAI(api_key=api)

    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "system", "content": "You are a doctor expert in drugs and diseases."},
            {"role": "user", "content": prompt}
        ])
  
    generated_text = response.choices[0].message.content

    return generated_text


# Define functions
def get_predictions(text):
    inputs = tokenizer(text, return_tensors="tf", truncation=True, max_length=512)
    outputs = model(inputs)
    probabilities = tf.nn.softmax(outputs.logits, axis=-1)
    predictions = []
    for idx, prob in enumerate(probabilities[0]):
        label = model.config.id2label[idx]
        predictions.append((label, prob.numpy()))
    return predictions

def analyze_probabilities(predictions):
    probabilities = [prob for _, prob in predictions]
    max_prob = max(probabilities)
    high_prob_label = next(label for label, prob in predictions if prob == max_prob)
    return  high_prob_label,max_prob


def main():
    # import symptoms data
    symptoms = get_data('symptomes.json')

    # Create sidebar and pages content
    tabs = ["Home", "About Us", "Model Details & Evaluations"]
    st.sidebar.header("Welcome to the Symptom checker app ! This app provide diagnosis based on your symptoms. Feel free to try it out !")
    st.sidebar.divider() 
    active_tab = st.sidebar.radio("Select Tab", tabs)
    if active_tab == "Home":
        st.header("Symptom Checker and Diagnosis App", divider = 'violet')
        image_path = "symptom.webp"
        col1, col2 = st.columns(2)
        with col1:
            st.image(image_path, use_column_width='auto')
        with col2:
            st.subheader("👩‍⚕️ Enter your symptoms:")
            st.subheader("get your diagnosis and useful advices")
  
        input_text = ""
        for category, symptoms in symptoms.items():
            st.sidebar.write(f"### {category}")
            category_symptoms = st.sidebar.multiselect(f"Select Symptoms in {category}", [symptom[0] for symptom in symptoms])
            if category_symptoms:
                input_text += f""
                for selected_symptom in category_symptoms:
                    symptom_description = next((symptom[1] for symptom in symptoms if symptom[0] == selected_symptom), "")
                    input_text += f"{symptom_description}\n"
   
        # Display the updated input_text
        manual_input=st.text_area("Symptoms", value=input_text, height=200)

        st.warning("If you want to know more, please enter your api key and click on submit: ")
        with st.expander("Click here to enter your api"):
            api=st.text_input("API KEY", value="",type='password')

        # Button to submit and get the predicted label
        if st.button("Submit"):
            predictions = get_predictions(manual_input)
            predicted_label,probability=analyze_probabilities(predictions)

            # Condition to display only high probability deseases
            if probability > 0.6:
                st.success(f"Based on your symptoms, there's a {100 * probability:.2f}% probability that you might have {predicted_label}.")

                if api:
     
                    # Try using the provided API key to call GPT-3
                    gpt3_prompt = f"Please provide a list of medications for {predicted_label}, along with a brief explanation of this disease."
                    #gpt3_response = call_gpt3(api, gpt3_prompt)

                    try :
                        gpt3_response = call_gpt3(api,gpt3_prompt)
                        # Display the GPT-3 response if successful
                        st.header(f'Information about the **{predicted_label}**')
                        st.info('Please note: This information is AI-generated and does not replace professional medical advice.')
                        st.write(gpt3_response)
  
                    except:
                    # Handle case where API call fails
                        st.warning("Unable to retrieve information using the provided API key. Please try another API key if available.")
                else:
                    st.write('If you want to know more, please enter your api key')
            else:
                st.warning("The symptoms you've described do not strongly indicate any of the 22 diseases in our database with a high probability. It's recommended to consult a healthcare professional for a more accurate diagnosis.")
                # Expander to show the list of diseases
                with st.expander("Click here to view the list of diseases"):
                    for disease in diseases:
                        st.write(disease)

    elif active_tab == "About Us":
        st.title("GZ-Health")
                
        st.markdown("**[Zahra ZABIHINPOUR](https://www.linkedin.com/in/zahra-zabihinpour/)**")
        st.markdown("**[Kevin GOUPIL](https://www.linkedin.com/in/kevin-goupil/)**")
        st.markdown(" We are a dynamic duo of data scientists collaborating to enhance our skills and stay at the forefront of the latest developments. With backgrounds in science and experience working with health data, we bring a unique blend of expertise to our data science projects. Our shared passion and commitment drive us to showcase and elevate our capabilities through innovative and impactful initiatives. Join us on this journey of continuous improvement and exploration in the world of data science. ")
        st.markdown(" ")

    elif active_tab == "Model Details & Evaluations":
        st.subheader("Model Overview:")
        st.write("This model is a fine-tuned adaptation of the bert-base-cased architecture, specifically designed for text classification tasks associated with diagnosing diseases based on symptoms. The primary goal is to scrutinize natural language symptom descriptions and accurately predict one of 22 potential diagnoses.")
        st.subheader("Dataset Information:")
        st.write("The model was trained on the Gretel/symptom_to_diagnosis dataset, which consists of 1,065 symptom descriptions in English, each labeled with one of the 22 possible diagnoses. This dataset focuses on detailed, fine-grained, single-domain diagnosis, making it suitable for tasks requiring nuanced symptom classification. For those interested in utilizing the model, the Symptom Checker and Diagnosis App, or the Inference API, are accessible at [https://huggingface.co/Zabihin/Symptom_to_Diagnosis](https://huggingface.co/Zabihin/Symptom_to_Diagnosis).")
        st.subheader("Model Performance Metrics:")
        metrics_data = pd.read_csv(io.StringIO(metrics_table), sep="|").dropna()
        st.table(metrics_data)



if __name__ == "__main__":
    main()