app.py

import gradio as gr
import openai
from openai import OpenAI
import google.generativeai as genai
import os
import io
import base64

# Set API key
#base_url = os.environ.get("OPENAI_API_BASE")
    
# Define the model to be used

DESCRIPTION = '''
<div>
<h1 style="text-align: center;">Medster - Medical Diagnostic Assistant</h1>
<p>An AI tool that helps you analyze symptoms and test reports.  </p>
<p>🔎 Select the department you need to consult, and enter the symptom description or physical examination information in the input box; you can also upload the test report image in the picture box.  </p>
<p>🦕 Please note that the generated information may be inaccurate and does not have any actual reference value. Please contact a professional doctor if necessary.  </p>
</div>
'''

css = """
h1 {
    text-align: center;
    display: block;
}
footer {
    display:none !important
}
"""

LICENSE = '[Medster](https://huggingface.co/spaces/vilarin/Medster)'

def read(filename):
    with open(filename) as f:
        data = f.read()
    return data

SYS_PROMPT = read('system_prompt.txt')

def endpoints(api_key):
    if api_key is not None:
        if api_key.startswith('sk-'):
            model_name = "gpt-4o"
            endpoint = 'OPENAI'
            return model_name, endpoint
        else:
            model_name = "models/gemini-1.5-pro-latest"
            endpoint = 'GOOGLE'
            return model_name, endpoint
            
def process_text(api_key, text_input, unit):
    model_name, endpoint = endpoints(api_key)
    if text_input and endpoint == 'OPENAI':
        client = OpenAI(api_key=api_key)
        completion = client.chat.completions.create(
            model=model_name,
            messages=[
                {"role": "system", "content": f" You are a experienced {unit} doctor." + SYS_PROMPT},
                {"role": "user", "content": f"Hello! Could you solve {text_input}?"}
            ]
        )
        return completion.choices[0].message.content
    elif text_input and endpoint == "GOOGLE":
        genai.configure(api_key=api_key)
        model = genai.GenerativeModel(model_name=model_name)
        prompt = f" You are a experienced {unit} doctor." + SYS_PROMPT + f"Could you solve {text_input}?"
        response = model.generate_content(prompt)
        return response.text
    return ""

def encode_image_to_base64(image_input):
    buffered = io.BytesIO()
    image_input.save(buffered, format="JPEG")
    img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
    return img_str

def process_image(api_key, image_input, unit):
    model_name, endpoint = endpoints(api_key)
    if image_input is not None and endpoint == 'OPENAI':
        client = OpenAI(api_key=api_key)
        #with open(image_input.name, "rb") as f:
        #    base64_image = base64.b64encode(f.read()).decode("utf-8")
        base64_image = encode_image_to_base64(image_input)
        response = client.chat.completions.create(
            model=model_name,
            messages=[
                {"role": "system", "content": f" You are a experienced {unit} doctor." + SYS_PROMPT},
                {"role": "user", "content": [
                    {"type": "text", "text": "Help me understand what is in this picture and analysis."},
                    {"type": "image_url", 
                     "image_url": {
                        "url": f"data:image/jpeg;base64,{base64_image}",
                        "detail":"low"}
                    }
                ]}
            ],
            temperature=0.0,
            max_tokens=1024,
        )
        return response.choices[0].message.content
    elif image_input is not None and endpoint == "GOOGLE":
        genai.configure(api_key=api_key)
        model = genai.GenerativeModel(model_name=model_name)
        prompt = f" You are a experienced {unit} doctor." + SYS_PROMPT + "Help me understand what is in this picture and analysis."
        response = model.generate_content([prompt, image_input],request_options={"timeout": 60})
        return response.text

def main(unit="", api_key="", text_input="", image_input=None):
    if text_input and image_input is None:
        return process_text(api_key, text_input, unit)
    elif image_input is not None:
        return process_image(api_key, image_input, unit)

with gr.Blocks(theme='shivi/calm_seafoam', css=css, title="Medster - Medical Diagnostic Assistant") as iface:
    with gr.Accordion(""):
        gr.Markdown(DESCRIPTION)
        unit = gr.Dropdown(label="🩺Department", value='Traditional Medicine', elem_id="units",
                             choices=["Traditional Medicine", "Internal Medicine", "Surgery", "Obstetrics and Gynecology", "Pediatrics", \
                                      "Orthodontics", "Andrology", "Dermatology and Venereology", "Infectious Diseases", "Psychiatry", \
                                         "Plastic Surgery Department", "Nutrition Department", "Reproductive Center", "Anesthesiology Department", "Medical Imaging Department", \
                                             "Orthopedics", "Oncology", "Emergency Department", "Laboratory Department"])
    with gr.Row():
        output_box = gr.Markdown(label="Diagnosis")  # Create an output textbox
    with gr.Row():
        api_key = gr.Textbox(label="API Key", type='password') # Input API key
    with gr.Row():
        image_input = gr.Image(type="pil", label="Upload Image")  # Create an image upload button
        text_input = gr.Textbox(label="Input Text")  # Create a text input box
    with gr.Row():
        submit_btn = gr.Button("🚀 Send")  # Create a submit button
        clear_btn = gr.ClearButton(output_box, value="🗑️ Clear") # Create a clear button

    # Set up the event listeners
    submit_btn.click(main, inputs=[unit, api_key, text_input, image_input], outputs=output_box)

    gr.Markdown(LICENSE)
    
#gr.close_all()

iface.queue().launch(show_api=False)  # Launch the Gradio interface