import streamlit as st import cv2 import numpy as np import tempfile import os import easyocr from PIL import Image, ImageDraw, ImageFont from deep_translator import GoogleTranslator import base64 from langchain.prompts import PromptTemplate from langchain.chains import LLMChain from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace os.environ["HUGGINGFACEHUB_API_KEY"] = os.getenv("HF") os.environ["HF_TOKEN"] = os.getenv("HF") st.set_page_config( page_title="MediAssist - Prescription Analyzer", layout="wide", page_icon="💊" ) def set_background(image_file): with open(image_file, "rb") as image: encoded = base64.b64encode(image.read()).decode() st.markdown( f""" """, unsafe_allow_html=True ) # Split large response into smaller chunks (for translation) def split_text_into_chunks(text, max_length=450): lines = text.split('\n') chunks = [] current = "" for line in lines: if len(current) + len(line) + 1 <= max_length: current += line + '\n' else: chunks.append(current.strip()) current = line + '\n' if current: chunks.append(current.strip()) return chunks def save_text_as_image(text, file_path): font = ImageFont.load_default() lines = text.split('\n') max_width = max([font.getbbox(line)[2] for line in lines]) + 20 line_height = font.getbbox(text)[3] + 10 img_height = line_height * len(lines) + 20 img = Image.new("RGB", (max_width, img_height), "white") draw = ImageDraw.Draw(img) y = 10 for line in lines: draw.text((10, y), line, font=font, fill="black") y += line_height img.save(file_path) return file_path set_background("background_img.jpg") # OCR @st.cache_resource def load_easyocr_reader(): return easyocr.Reader(['en']) st.sidebar.title("💊 MediAssist") st.sidebar.markdown("Analyze prescriptions with ease using AI") st.sidebar.markdown("---") st.sidebar.markdown("🔗 **Connect with me:**") st.sidebar.markdown("""

""", unsafe_allow_html=True) st.sidebar.markdown("---") st.markdown("""

🧠 MediAssist

Prescription Analyzer using AI and OCR

Upload a doctor's prescription image, and MediAssist will extract, translate, and explain it for you.

""", unsafe_allow_html=True) uploaded_file = st.file_uploader("📤 Upload Prescription Image (JPG/PNG)", type=["jpg", "jpeg", "png"]) if uploaded_file: with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file: temp_file.write(uploaded_file.read()) orig_path = temp_file.name # Image preprocessing image = cv2.imread(orig_path) gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) _, binary_inv = cv2.threshold(gray, 128, 255, cv2.THRESH_BINARY_INV) kernel = np.ones((3, 3), np.uint8) dilated = cv2.dilate(binary_inv, kernel, iterations=1) dilated_path = orig_path.replace(".png", "_dilated.png") cv2.imwrite(dilated_path, dilated) reader = load_easyocr_reader() text_list = reader.readtext(dilated, detail=0) text = "\n".join(text_list) col1, col2 = st.columns([1, 2]) with col1: st.image(dilated, caption="🧾 Preprocessed Prescription", channels="GRAY", use_container_width=True) with col2: st.success("✅ Image Uploaded and Preprocessed") st.markdown("#### 📝 Extracted Text") st.code(text) # Prompt LLM template = """ You are a helpful and structured medical assistant. Below is a prescription text extracted from an image: {prescription_text} Your tasks: 1. Identify and list only the **medicine names** mentioned (ignore other irrelevant text). 2. For each identified medicine, provide the following: - Dosage and Timing - Possible Side Effects - Special Instructions 🧾 Format your response clearly and neatly as follows: - Medicine Name 1 - Dosage and Timing: ... - Side Effects: ... - Special Instructions: ... - Medicine Name 2 - Dosage and Timing: ... - Side Effects: ... - Special Instructions: ... Ensure each medicine starts with a new bullet point and all details are on separate lines. """ prompt = PromptTemplate(input_variables=["prescription_text"], template=template) llm_model = HuggingFaceEndpoint( repo_id="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", provider="novita", temperature=0.6, max_new_tokens=300, task="conversational" ) llm = ChatHuggingFace( llm=llm_model, repo_id="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", provider="novita", temperature=0.6, max_new_tokens=300, task="conversational" ) chain = LLMChain(llm=llm, prompt=prompt) response = "" hindi_text = "" if st.button("🔍 Analyze Extracted Text"): with st.spinner("Analyzing with LLM..."): response = chain.run(prescription_text=text) st.markdown("#### 💡 AI-based Medicine Analysis") st.text_area("LLM Output", response, height=600) # Save txt and image txt_path = "medicine_analysis.txt" with open(txt_path, "w") as f: f.write(response) img_path = "medicine_analysis.png" save_text_as_image(response, img_path) st.markdown("#### 📥 Download (English)") col1, col2 = st.columns(2) with col1: st.download_button("⬇️ English TXT", data=response.encode(), file_name="medicine_analysis.txt") with col2: with open(img_path, "rb") as img_file: st.download_button("🖼️ English Image", data=img_file, file_name="medicine_analysis.png", mime="image/png")\ if response and st.button("🌐 Translate to Hindi"): with st.spinner("Translating to Hindi..."): chunks = split_text_into_chunks(response, max_length=100) hindi_chunks = [] for chunk in chunks: try: translated = GoogleTranslator(source='auto', target='hi').translate(chunk) hindi_chunks.append(translated) except Exception as e: hindi_chunks.append("[Error translating chunk]") hindi_text = "\n\n".join(hindi_chunks) st.markdown("#### 🌐 Hindi Translation") st.text_area("Translated Output (Hindi)", hindi_text, height=600) hindi_img_path = "hindi_output.png" save_text_as_image(hindi_text, hindi_img_path) st.markdown("#### 📥 Download (Hindi)") col3, col4 = st.columns(2) with col3: st.download_button("⬇️ Hindi TXT", data=hindi_text.encode(), file_name="hindi_medicine_analysis.txt") with col4: with open(hindi_img_path, "rb") as img_file: st.download_button("🖼️ Hindi Image", data=img_file, file_name="hindi_medicine_analysis.png", mime="image/png") try: os.remove(orig_path) os.remove(dilated_path) except: pass else: st.markdown("📸 Please Upload Scanned prescription image to get best result", unsafe_allow_html=True) # import streamlit as st # import cv2 # import numpy as np # import tempfile # import os # import easyocr # from langchain.prompts import PromptTemplate # from langchain.chains import LLMChain # from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace # # Set Hugging Face API keys # os.environ["HUGGINGFACEHUB_API_KEY"] = os.getenv("HF") # os.environ["HF_TOKEN"] = os.getenv("HF") # # Streamlit page setup # st.set_page_config( # page_title="MediAssist - Prescription Analyzer", # layout="wide", # page_icon="💊" # ) # st.sidebar.title("💊 MediAssist") # st.sidebar.markdown("Analyze prescriptions with ease using AI") # st.sidebar.markdown("---") # st.sidebar.markdown("🔗 **Connect with me:**") # st.sidebar.markdown(""" #

# #

# # #

# #

# """, unsafe_allow_html=True) # st.sidebar.markdown("---") # st.markdown(""" #

🧠 MediAssist

Prescription Analyzer using AI and OCR

Upload a doctor's prescription image, and MediAssist will extract, translate, and explain it for you.

#
# """, unsafe_allow_html=True) # uploaded_file = st.file_uploader("📤 Upload Prescription Image (JPG/PNG)", type=["jpg", "jpeg", "png"]) # if uploaded_file: # with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file: # temp_file.write(uploaded_file.read()) # orig_path = temp_file.name # # Preprocessing # image = cv2.imread(orig_path) # gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # _, binary_inv = cv2.threshold(gray, 128, 255, cv2.THRESH_BINARY_INV) # kernel = np.ones((3, 3), np.uint8) # dilated = cv2.dilate(binary_inv, kernel, iterations=1) # # Save preprocessed image for future reference/removal # dilated_path = orig_path.replace(".png", "_dilated.png") # cv2.imwrite(dilated_path, dilated) # # OCR using EasyOCR # reader = easyocr.Reader(['en']) # text_list = reader.readtext(dilated, detail=0) # text = "\n".join(text_list) # # Prompt Template # template = """ # You are a helpful medical assistant. # Here is a prescription text extracted from an image: # {prescription_text} # Please do the following: # 1. Extract only the medicine names mentioned in the prescription (ignore any other text). # 2. For each medicine, provide: # - When to take it (timing and dosage) # - Possible side effects # - Any special instructions # Format your answer as bullet points, listing only medicines and their details. # """ # prompt = PromptTemplate(input_variables=["prescription_text"], template=template) # llm_model = HuggingFaceEndpoint( # repo_id="aaditya/Llama3-OpenBioLLM-70B", # provider="nebius", # temperature=0.6, # max_new_tokens=300, # task="conversational" # ) # llm = ChatHuggingFace( # llm=llm_model, # repo_id="aaditya/Llama3-OpenBioLLM-70B", # provider="nebius", # temperature=0.6, # max_new_tokens=300, # task="conversational" # ) # chain = LLMChain(llm=llm, prompt=prompt) # col1, col2 = st.columns([1, 2]) # with col1: # st.image(dilated, caption="Preprocessed Prescription", channels="GRAY", use_container_width=True) # with col2: # st.success("✅ Prescription Uploaded & Preprocessed Successfully") # st.markdown("### 📜 Extracted Text") # st.code(text) # if st.button("🔍 Analyze Text"): # with st.spinner("Analyzing..."): # response = chain.run(prescription_text=text) # st.success(response) # # Cleanup temp files # os.remove(orig_path) # os.remove(dilated_path) # else: # st.markdown("Upload a prescription image to begin analysis.", unsafe_allow_html=True) # import streamlit as st # import cv2 # import numpy as np # import tempfile # import os # # import pytesseract # import easyocr # # from langchain.document_loaders.image import UnstructuredImageLoader # # from langchain_community.document_loaders import UnstructuredImageLoader # from langchain.prompts import PromptTemplate # from langchain.chains import LLMChain # from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace # # Set Hugging Face API keys # os.environ["HUGGINGFACEHUB_API_KEY"] = os.getenv("HF") # os.environ["HF_TOKEN"] = os.getenv("HF") # st.set_page_config( # page_title="MediAssist - Prescription Analyzer", # layout="wide", # page_icon="💊" # ) # st.sidebar.title("💊 MediAssist") # st.sidebar.markdown("Analyze prescriptions with ease using AI") # st.sidebar.markdown("---") # st.sidebar.markdown("🔗 **Connect with me:**") # st.sidebar.markdown(""" #

# #

# # #

# #

# """, unsafe_allow_html=True) # st.sidebar.markdown("---") # st.markdown(""" #

🧠 MediAssist

Prescription Analyzer using AI and OCR

Upload a doctor's prescription image, and MediAssist will extract, translate, and explain it for you.

#
# """, unsafe_allow_html=True) # uploaded_file = st.file_uploader("📤 Upload Prescription Image (JPG/PNG)", type=["jpg", "jpeg", "png"]) # if uploaded_file: # with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file: # temp_file.write(uploaded_file.read()) # orig_path = temp_file.name # # Step 1: Read and preprocess image # image = cv2.imread(orig_path) # gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # _, binary_inv = cv2.threshold(gray, 128, 255, cv2.THRESH_BINARY_INV) # kernel = np.ones((3, 3), np.uint8) # dilated = cv2.dilate(binary_inv, kernel, iterations=1) # reader = easyocr.Reader(['en']) # text_list = reader.readtext(dilated, detail=0) # text = "\n".join(text_list) # # text = pytesseract.image_to_string(dilated) # # Save preprocessed image for OCR # # dilated_path = orig_path.replace(".png", "_dilated.png") # # cv2.imwrite(dilated_path, dilated) # # loader = UnstructuredImageLoader(dilated_path) # # documents = loader.load() # # extracted_text = "\n".join([doc.page_content for doc in documents]) # template = """ # You are a helpful medical assistant. # Here is a prescription text extracted from an image: # {prescription_text} # Please do the following: # 1. Extract only the medicine names mentioned in the prescription (ignore any other text). # 2. For each medicine, provide: # - When to take it (timing and dosage) # - Possible side effects # - Any special instructions # Format your answer as bullet points, listing only medicines and their details. # """ # prompt = PromptTemplate(input_variables=["prescription_text"], template=template) # llm_model = HuggingFaceEndpoint( # repo_id="aaditya/Llama3-OpenBioLLM-70B", # provider="nebius", # temperature=0.6, # max_new_tokens=300, # task="conversational" # ) # model = ChatHuggingFace( # llm=llm_model, # repo_id="aaditya/Llama3-OpenBioLLM-70B", # provider="nebius", # temperature=0.6, # max_new_tokens=300, # task="conversational" # ) # chain = LLMChain(llm=model, prompt=prompt) # col1, col2 = st.columns([1, 2]) # with col1: # st.image(dilated, caption="Preprocessed Prescription", channels="GRAY", use_container_width=True) # with col2: # st.success("✅ Prescription Uploaded & Preprocessed Successfully") # st.markdown("### 📜 Extracted Text") # st.code(text) # # st.code(extracted_text) # if st.button("🔍 Analyze Text"): # with st.spinner("Analyzing..."): # response = chain.run(prescription_text=text) # # response = chain.run(prescription_text=extracted_text) # st.success(response) # # Cleanup temp files # os.remove(orig_path) # os.remove(dilated_path) # else: # st.markdown("Upload a prescription image to begin analysis.", unsafe_allow_html=True) # st.markdown("### 🌐 Translated Text") # st.code("पेरासिटामोल 500 मिलीग्राम\nभोजन के बाद दिन में दो बार 1 गोली लें", language='text') # st.markdown("### ⏱️ Tablet Timing & Instructions") # st.info("- Morning after breakfast\n- Night after dinner\n- Take with water\n- Do not exceed 2 tablets in 24 hours") # st.markdown("### ⚠️ Possible Side Effects") # st.warning("- Nausea\n- Dizziness\n- Liver damage (on overdose)") # os.remove(temp_path) # os.remove(orig_path) # os.remove(dilated_path)