Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,133 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import subprocess
|
| 3 |
+
from fastapi import FastAPI
|
| 4 |
+
from sentence_transformers import SentenceTransformer
|
| 5 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 6 |
+
import fitz # PyMuPDF pour extraction du texte PDF
|
| 7 |
+
import faiss
|
| 8 |
+
import numpy as np
|
| 9 |
+
import requests
|
| 10 |
+
import pytesseract
|
| 11 |
+
from PIL import Image
|
| 12 |
+
import gradio as gr
|
| 13 |
+
|
| 14 |
+
# Installation automatique de Tesseract pour Windows
|
| 15 |
+
def install_tesseract_windows():
|
| 16 |
+
tesseract_installer_url = "https://github.com/tesseract-ocr/tesseract/releases/download/5.3.0/tesseract-5.3.0.20221214.exe"
|
| 17 |
+
installer_path = "tesseract_installer.exe"
|
| 18 |
+
|
| 19 |
+
if not os.path.exists("C:\\Program Files\\Tesseract-OCR"):
|
| 20 |
+
print("Installation de Tesseract OCR...")
|
| 21 |
+
# Télécharger l'installateur
|
| 22 |
+
subprocess.run(["curl", "-L", tesseract_installer_url, "-o", installer_path], shell=True, check=True)
|
| 23 |
+
# Exécuter l'installateur silencieusement
|
| 24 |
+
subprocess.run([installer_path, "/S"], shell=True, check=True)
|
| 25 |
+
os.environ["TESSDATA_PREFIX"] = "C:\\Program Files\\Tesseract-OCR\\"
|
| 26 |
+
pytesseract.pytesseract.tesseract_cmd = "C:\\Program Files\\Tesseract-OCR\\tesseract.exe"
|
| 27 |
+
print("Tesseract OCR installé avec succès.")
|
| 28 |
+
else:
|
| 29 |
+
pytesseract.pytesseract.tesseract_cmd = "C:\\Program Files\\Tesseract-OCR\\tesseract.exe"
|
| 30 |
+
print("Tesseract OCR déjà installé.")
|
| 31 |
+
|
| 32 |
+
install_tesseract_windows()
|
| 33 |
+
|
| 34 |
+
# Configuration de l'API Gemini
|
| 35 |
+
GEMINI_API_KEY = "AIzaSyArbgg_p_HlmpgrcjVYemdSJeMCP9OTj3E"
|
| 36 |
+
GEMINI_API_URL = "https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent"
|
| 37 |
+
|
| 38 |
+
# Configuration FAISS et embeddings
|
| 39 |
+
EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
|
| 40 |
+
model = SentenceTransformer(EMBEDDING_MODEL)
|
| 41 |
+
INDEX_PATH = "medical_faiss_index"
|
| 42 |
+
documents = []
|
| 43 |
+
|
| 44 |
+
if os.path.exists(INDEX_PATH):
|
| 45 |
+
index = faiss.read_index(INDEX_PATH)
|
| 46 |
+
print("Index FAISS chargé avec succès.")
|
| 47 |
+
else:
|
| 48 |
+
index = faiss.IndexFlatL2(model.get_sentence_embedding_dimension())
|
| 49 |
+
print("Nouvel index FAISS créé.")
|
| 50 |
+
|
| 51 |
+
# Fonction d'extraction de texte
|
| 52 |
+
def extract_text_from_pdf(file_content: bytes):
|
| 53 |
+
pdf = fitz.open(stream=file_content, filetype="pdf")
|
| 54 |
+
paragraphs = []
|
| 55 |
+
for page in pdf:
|
| 56 |
+
text = page.get_text()
|
| 57 |
+
if text.strip():
|
| 58 |
+
paragraphs.extend([p.strip() for p in text.split("\n\n") if p.strip()])
|
| 59 |
+
return paragraphs
|
| 60 |
+
|
| 61 |
+
def extract_text_from_image(file_path):
|
| 62 |
+
return pytesseract.image_to_string(Image.open(file_path))
|
| 63 |
+
|
| 64 |
+
# Fonction pour ajouter les documents de référence
|
| 65 |
+
def add_medical_reference(file):
|
| 66 |
+
with open(file.name, "rb") as f:
|
| 67 |
+
file_content = f.read()
|
| 68 |
+
paragraphs = extract_text_from_pdf(file_content)
|
| 69 |
+
embeddings = model.encode(paragraphs)
|
| 70 |
+
index.add(np.array(embeddings, dtype="float32"))
|
| 71 |
+
documents.extend(paragraphs)
|
| 72 |
+
faiss.write_index(index, INDEX_PATH)
|
| 73 |
+
return "Référence médicale ajoutée avec succès."
|
| 74 |
+
|
| 75 |
+
# Fonction pour analyser un fichier (PDF ou Image)
|
| 76 |
+
def analyze_blood_test(file):
|
| 77 |
+
try:
|
| 78 |
+
if file.name.endswith((".png", ".jpg", ".jpeg")):
|
| 79 |
+
extracted_text = extract_text_from_image(file.name)
|
| 80 |
+
elif file.name.endswith(".pdf"):
|
| 81 |
+
with open(file.name, "rb") as f:
|
| 82 |
+
file_content = f.read()
|
| 83 |
+
paragraphs = extract_text_from_pdf(file_content)
|
| 84 |
+
extracted_text = "\n".join(paragraphs)
|
| 85 |
+
else:
|
| 86 |
+
return "Format non supporté. Utilisez un PDF ou une image."
|
| 87 |
+
|
| 88 |
+
if not extracted_text.strip():
|
| 89 |
+
return "Aucun texte valide extrait."
|
| 90 |
+
|
| 91 |
+
# Recherche et appel Gemini
|
| 92 |
+
relevant_docs = search_faiss(extracted_text, k=5)
|
| 93 |
+
context = "\n".join(relevant_docs)
|
| 94 |
+
enriched_prompt = f"Voici les résultats d'analyse :\n{extracted_text}\n\nContexte pertinent :\n{context}"
|
| 95 |
+
gemini_response = call_gemini_api(enriched_prompt)
|
| 96 |
+
|
| 97 |
+
return f"Réponse générée :\n{gemini_response}\n\nDocuments pertinents :\n{relevant_docs}"
|
| 98 |
+
except Exception as e:
|
| 99 |
+
return f"Erreur : {str(e)}"
|
| 100 |
+
|
| 101 |
+
# Recherche FAISS
|
| 102 |
+
def search_faiss(query, k=5):
|
| 103 |
+
query_embedding = model.encode([query])
|
| 104 |
+
distances, indices = index.search(np.array(query_embedding, dtype="float32"), k)
|
| 105 |
+
return [documents[i] for i in indices[0] if i < len(documents)]
|
| 106 |
+
|
| 107 |
+
# Appel API Gemini
|
| 108 |
+
def call_gemini_api(prompt):
|
| 109 |
+
headers = {"Content-Type": "application/json"}
|
| 110 |
+
payload = {"contents": [{"parts": [{"text": prompt}]}]}
|
| 111 |
+
try:
|
| 112 |
+
response = requests.post(f"{GEMINI_API_URL}?key={GEMINI_API_KEY}", json=payload, headers=headers)
|
| 113 |
+
return response.json().get("candidates", [{}])[0].get("content", {}).get("parts", [{}])[0].get("text", "Pas de réponse.")
|
| 114 |
+
except Exception as e:
|
| 115 |
+
return f"Erreur API : {str(e)}"
|
| 116 |
+
|
| 117 |
+
# Interface Gradio
|
| 118 |
+
with gr.Blocks() as demo:
|
| 119 |
+
gr.Markdown("## Analyse Médicale avec RAG et Gemini")
|
| 120 |
+
|
| 121 |
+
with gr.Tab("Ajouter Références Médicales"):
|
| 122 |
+
ref_file = gr.File(label="Téléchargez un fichier PDF de référence médicale")
|
| 123 |
+
ref_output = gr.Textbox(label="Résultat")
|
| 124 |
+
ref_button = gr.Button("Ajouter")
|
| 125 |
+
ref_button.click(add_medical_reference, inputs=ref_file, outputs=ref_output)
|
| 126 |
+
|
| 127 |
+
with gr.Tab("Analyser un Résultat d'Analyse"):
|
| 128 |
+
test_file = gr.File(label="Téléchargez un fichier PDF ou image (JPG/PNG)")
|
| 129 |
+
analysis_output = gr.Textbox(label="Résultat")
|
| 130 |
+
analyze_button = gr.Button("Analyser")
|
| 131 |
+
analyze_button.click(analyze_blood_test, inputs=test_file, outputs=analysis_output)
|
| 132 |
+
|
| 133 |
+
demo.launch()
|