Update app.py

app.py

@@ -1,7 +1,4 @@
-# Install required libraries
-#!pip install opencv-python-headless pytesseract gradio
-#!apt-get update && apt-get install -y tesseract-ocr
-
+import os
 import cv2
 import numpy as np
 import pytesseract
@@ -11,167 +8,195 @@ import base64
 from datetime import datetime
 import pytz
 import urllib.parse
+import re
+
+# Install Tesseract OCR in the Hugging Face container
+os.system("apt-get update && apt-get install -y tesseract-ocr")
 
 # Set timezone to IST
 ist = pytz.timezone('Asia/Kolkata')
 current_time = datetime.now(ist).strftime("%Y-%m-%d %I:%M %p IST")
 
+# Function to validate phone number (e.g., +91 followed by 10 digits)
+def validate_phone(phone):
+    pattern = r"^\+\d{1,3}\d{10}$"
+    return bool(re.match(pattern, phone))
+
 # Function to preprocess poor-quality images
 def preprocess_image(image):
-    # Convert to grayscale
-    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
-    
-    # Denoise the image using Gaussian blur
-    denoised = cv2.GaussianBlur(gray, (5, 5), 0)
-    
-    # Sharpen the image using a kernel
-    sharpening_kernel = np.array([[-1, -1, -1],
-                                  [-1,  9, -1],
-                                  [-1, -1, -1]])
-    sharpened = cv2.filter2D(denoised, -1, sharpening_kernel)
-    
-    # Adjust contrast using CLAHE (Contrast Limited Adaptive Histogram Equalization)
-    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
-    contrast_adjusted = clahe.apply(sharpened)
-    
-    return contrast_adjusted
+    try:
+        # Convert to grayscale
+        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+        
+        # Denoise the image using Gaussian blur
+        denoised = cv2.GaussianBlur(gray, (5, 5), 0)
+        
+        # Sharpen the image using a kernel
+        sharpening_kernel = np.array([[-1, -1, -1],
+                                     [-1,  9, -1],
+                                     [-1, -1, -1]])
+        sharpened = cv2.filter2D(denoised, -1, sharpening_kernel)
+        
+        # Adjust contrast using CLAHE
+        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+        contrast_adjusted = clahe.apply(sharpened)
+        
+        return contrast_adjusted, None
+    except Exception as e:
+        return None, f"Error preprocessing image: {str(e)}"
 
 # Function to analyze the eye image
 def analyze_eye_image(image, patient_name, doctor_email, doctor_phone):
-    # Convert Gradio image (numpy array) to OpenCV format
-    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
-    
-    # Keep a copy of the original image for annotations
-    annotated_image = image.copy()
-    
-    # Preprocess the image to handle poor quality
-    processed_image = preprocess_image(image)
-    
-    # Apply adaptive thresholding to improve text detection for poor-quality images
-    thresh = cv2.adaptiveThreshold(processed_image, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
-                                   cv2.THRESH_BINARY_INV, 11, 2)
-    
-    # Detect text using pytesseract
-    text_data = pytesseract.image_to_data(thresh, output_type=pytesseract.Output.DICT)
-    
-    # Store detected labels and their positions
-    labels = []
-    for i in range(len(text_data['text'])):
-        confidence = int(text_data['conf'][i])
-        if confidence > 50:  # Lowered confidence threshold for poor-quality images
-            label = text_data['text'][i].strip()
-            if label:  # Ignore empty strings
-                x, y = text_data['left'][i], text_data['top'][i]
-                w, h = text_data['width'][i], text_data['height'][i]
-                labels.append({'label': label, 'position': (x, y), 'size': (w, h)})
-                
-                # Annotate the image with a bounding box and label
-                cv2.rectangle(annotated_image, (x, y), (x + w, y + h), (0, 255, 0), 2)
-                cv2.putText(annotated_image, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
-    
-    # Merge "Optic" and "nerve" if they are close to each other
-    merged_labels = []
-    i = 0
-    while i < len(labels):
-        if i + 1 < len(labels) and labels[i]['label'].lower() == 'optic' and labels[i + 1]['label'].lower() == 'nerve':
-            merged_label = "Optic Nerve"
-            x, y = labels[i]['position']
-            merged_labels.append({'label': merged_label, 'position': (x, y)})
-            i += 2
+    try:
+        # Input validation
+        if image is None or not isinstance(image, np.ndarray) or image.size == 0:
+            return None, "Error: No valid image uploaded. Please upload a PNG or JPEG image.", "", "", ""
+        if not patient_name.strip():
+            return None, "Error: Patient name is required.", "", "", ""
+        if not doctor_email.strip():
+            return None, "Error: Doctor's email is required.", "", "", ""
+        if not doctor_phone.strip() or not validate_phone(doctor_phone):
+            return None, "Error: Invalid phone number. Use format: +91XXXXXXXXXX", "", "", ""
+
+        # Convert Gradio image (numpy) to OpenCV format
+        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+        annotated_image = image.copy()
+
+        # Preprocess the image
+        processed_image, error = preprocess_image(image)
+        if error:
+            return None, error, "", "", ""
+
+        # Apply adaptive thresholding
+        thresh = cv2.adaptiveThreshold(processed_image, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
+                                      cv2.THRESH_BINARY_INV, 11, 2)
+
+        # Detect text using pytesseract
+        text_data = pytesseract.image_to_data(thresh, output_type=pytesseract.Output.DICT)
+
+        # Store detected labels and their positions
+        labels = []
+        for i in range(len(text_data['text'])):
+            confidence = int(text_data['conf'][i])
+            if confidence > 50:  # Lowered threshold for poor-quality images
+                label = text_data['text'][i].strip()
+                if label:
+                    x, y = text_data['left'][i], text_data['top'][i]
+                    w, h = text_data['width'][i], text_data['height'][i]
+                    labels.append({'label': label, 'position': (x, y), 'size': (w, h)})
+                    # Annotate the image
+                    cv2.rectangle(annotated_image, (x, y), (x + w, y + h), (0, 255, 0), 2)
+                    cv2.putText(annotated_image, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
+
+        # Merge "Optic" and "nerve" if close
+        merged_labels = []
+        i = 0
+        while i < len(labels):
+            if i + 1 < len(labels) and labels[i]['label'].lower() == 'optic' and labels[i + 1]['label'].lower() == 'nerve':
+                merged_label = "Optic Nerve"
+                x, y = labels[i]['position']
+                merged_labels.append({'label': merged_label, 'position': (x, y)})
+                i += 2
+            else:
+                merged_labels.append(labels[i])
+                i += 1
+
+        # Generate report
+        report = f"Eye Image Analysis Report for {patient_name} (Generated on {current_time}):\n\n"
+        report += "Detected Text and Their Positions:\n"
+        if merged_labels:
+            for item in merged_labels:
+                report += f"- {item['label']} at position (x: {item['position'][0]}, y: {item['position'][1]})\n"
+        else:
+            report += "No text detected in the image.\n"
+
+        # Image Quality Assessment
+        report += "\nImage Quality Assessment:\n"
+        sharpness = cv2.Laplacian(processed_image, cv2.CV_64F).var()
+        if sharpness < 50:
+            report += "- The image quality is poor (blurry or low resolution). This may affect analysis accuracy.\n"
+            report += "- Recommendation: Use a higher-quality image or improve lighting and focus.\n"
         else:
-            merged_labels.append(labels[i])
-            i += 1
-    
-    # Generate a detailed analysis report
-    report = f"Eye Image Analysis Report for {patient_name} (Generated on {current_time}):\n\n"
-    report += "Detected Text and Their Positions:\n"
-    if merged_labels:
-        for item in merged_labels:
-            report += f"- {item['label']} at position (x: {item['position'][0]}, y: {item['position'][1]})\n"
-    else:
-        report += "No text detected in the image.\n"
-    
-    # Image Quality Assessment
-    report += "\nImage Quality Assessment:\n"
-    # Estimate image quality based on sharpness (using Laplacian variance)
-    sharpness = cv2.Laplacian(processed_image, cv2.CV_64F).var()
-    if sharpness < 50:
-        report += "- The image quality appears to be poor (blurry or low resolution). This may affect the accuracy of the analysis.\n"
-        report += "- Recommendation: Use a higher-quality image or improve lighting and focus during image capture.\n"
-    else:
-        report += "- The image quality appears to be sufficient for analysis.\n"
-    
-    # Detailed Analysis
-    report += "\nDetailed Analysis:\n"
-    detected_labels = [item['label'] for item in merged_labels]
-    if "iStock" in detected_labels and "Credit" in detected_labels and "Gannet77" in detected_labels:
-        report += "- The detected text ('iStock', 'Credit', 'Gannet77') suggests that this image is likely a stock image from iStock, credited to the contributor 'Gannet77'.\n"
-        report += "- This image does not appear to be an actual patient eye scan (e.g., fundus image). It might be a labeled diagram or an illustrative image.\n"
-        report += "- No anatomical features (e.g., optic disc, macula, blood vessels) were detected in this analysis, likely because the image is not a real eye scan.\n"
-    else:
-        report += "- No specific anatomical features were detected. This could be due to the image quality, type, or limitations of the text detection system.\n"
-        report += "- For poor-quality images, consider using a fundus camera or smartphone attachment with better focus and lighting.\n"
-    
-    # Preliminary Abnormality Check
-    report += "\nPreliminary Abnormality Check:\n"
-    report += "- This is a placeholder. Integrate a deep learning model to detect abnormalities like diabetic retinopathy or glaucoma.\n"
-    report += "- Since this appears to be a stock image, no actual abnormality detection is applicable.\n"
-    report += "- For poor-quality images, abnormality detection accuracy may be reduced. A higher-quality image is recommended.\n"
-    
-    # Recommendations for the Doctor
-    report += "\nRecommendations for the Doctor:\n"
-    report += "- Request the patient to provide a real eye scan (e.g., a fundus image) using a fundus camera or smartphone attachment for accurate analysis.\n"
-    report += "- If this is intended for educational purposes, confirm the anatomical labels manually and use the image for reference.\n"
-    report += "- For clinical diagnosis, a high-quality fundus image or slit-lamp image is recommended to assess conditions like diabetic retinopathy, glaucoma, or cataracts.\n"
-    if sharpness < 50:
-        report += "- The image quality is poor. Advise the patient to recapture the image with better lighting, focus, and resolution.\n"
-    
-    # Convert annotated image back to RGB for Gradio display
-    annotated_image = cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
-    
-    # Convert the report to a downloadable text file
-    report_file = io.StringIO(report)
-    report_bytes = report.encode('utf-8')
-    report_b64 = base64.b64encode(report_bytes).decode('utf-8')
-    report_download_link = f'<a href="data:text/plain;base64,{report_b64}" download="eye_analysis_report_{patient_name}.txt">Download Report</a>'
-    
-    # Generate WhatsApp link
-    whatsapp_message = f"Eye Image Analysis Report for {patient_name}:\n\n{report}"
-    whatsapp_link = f"https://wa.me/{doctor_phone}?text={urllib.parse.quote(whatsapp_message)}"
-    whatsapp_html = f'<a href="{whatsapp_link}" target="_blank">Send Report via WhatsApp</a>'
-    
-    # Simulate email content (to be sent to the doctor)
-    email_content = f"Subject: Eye Image Analysis Report for {patient_name}\n\n"
-    email_content += "Dear Doctor,\n\n"
-    email_content += f"Please find the analysis report for patient {patient_name} below:\n\n"
-    email_content += report
-    email_content += "\nThe annotated image and report are attached for your review.\n"
-    email_content += "For further details, please contact the patient or the system administrator.\n\n"
-    email_content += "Best regards,\nEye Analysis System\n\n"
-    email_content += f"Note: To send this email to {doctor_email}, deploy this app on a server with email functionality (e.g., using smtplib in Python)."
-
-    return annotated_image, report, report_download_link, email_content, whatsapp_html
-
-# Gradio interface for patient use
+            report += "- The image quality is sufficient for analysis.\n"
+
+        # Detailed Analysis
+        report += "\nDetailed Analysis:\n"
+        detected_labels = [item['label'] for item in merged_labels]
+        if "iStock" in detected_labels and "Credit" in detected_labels and "Gannet77" in detected_labels:
+            report += "- Detected text ('iStock', 'Credit', 'Gannet77') suggests this is a stock image from iStock, credited to 'Gannet77'.\n"
+            report += "- This is likely not a real patient eye scan but a labeled diagram.\n"
+            report += "- No anatomical features (e.g., optic disc, macula) detected, as this is not a real scan.\n"
+        else:
+            report += "- No specific anatomical features detected, possibly due to image quality or type.\n"
+            report += "- For poor-quality images, use a fundus camera or smartphone attachment.\n"
+
+        # Preliminary Abnormality Check
+        report += "\nPreliminary Abnormality Check:\n"
+        report += "- Placeholder: Integrate a deep learning model for abnormality detection (e.g., diabetic retinopathy).\n"
+        report += "- For stock images, abnormality detection is not applicable.\n"
+        if sharpness < 50:
+            report += "- Poor image quality may reduce abnormality detection accuracy.\n"
+
+        # Recommendations
+        report += "\nRecommendations for the Doctor:\n"
+        report += "- Request a real eye scan (e.g., fundus image) for accurate analysis.\n"
+        report += "- For educational use, confirm anatomical labels manually.\n"
+        report += "- For clinical diagnosis, use high-quality fundus or slit-lamp images.\n"
+        if sharpness < 50:
+            report += "- Advise patient to recapture image with better lighting and resolution.\n"
+
+        # Convert annotated image back to RGB
+        annotated_image = cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
+
+        # Create downloadable report
+        report_bytes = report.encode('utf-8')
+        report_b64 = base64.b64encode(report_bytes).decode('utf-8')
+        report_download_link = f'<a href="data:text/plain;base64,{report_b64}" download="eye_analysis_report_{patient_name}.txt">Download Report</a>'
+
+        # Generate WhatsApp link
+        whatsapp_message = f"Eye Image Analysis Report for {patient_name}:\n\n{report}"
+        whatsapp_link = f"https://wa.me/{doctor_phone}?text={urllib.parse.quote(whatsapp_message)}"
+        whatsapp_html = f'<a href="{whatsapp_link}" target="_blank">Send Report via WhatsApp</a>'
+
+        # Simulate email content
+        email_content = f"Subject: Eye Image Analysis Report for {patient_name}\n\n"
+        email_content += "Dear Doctor,\n\n"
+        email_content += f"Analysis report for patient {patient_name}:\n\n{report}"
+        email_content += "\nThe annotated image and report are attached for review.\n"
+        email_content += "Note: To send this email, deploy with email functionality (e.g., smtplib).\n"
+
+        return annotated_image, report, report_download_link, email_content, whatsapp_html
+
+    except Exception as e:
+        return None, f"Error processing image: {str(e)}", "", "", ""
+
+# Gradio interface
 interface = gr.Interface(
-    fn=analyze_eye_image,  # Function to call
+    fn=analyze_eye_image,
     inputs=[
-        gr.Image(label="Upload Eye Image"),  # Input: eye image
-        gr.Textbox(label="Patient Name", placeholder="Enter patient's name"),  # Input: patient name
-        gr.Textbox(label="Doctor's Email", placeholder="Enter doctor's email address"),  # Input: doctor's email
-        gr.Textbox(label="Doctor's Phone Number (WhatsApp)", placeholder="Enter doctor's phone number (e.g., +919876543210)")  # Input: doctor's phone number
+        gr.Image(label="Upload Eye Image (PNG/JPEG)", type="numpy"),
+        gr.Textbox(label="Patient Name", placeholder="Enter patient's name"),
+        gr.Textbox(label="Doctor's Email", placeholder="Enter doctor's email (e.g., doctor@example.com)"),
+        gr.Textbox(label="Doctor's Phone Number (WhatsApp)", placeholder="Enter phone number (e.g., +919876543210)")
     ],
     outputs=[
-        gr.Image(label="Annotated Image"),  # Output: annotated image
-        gr.Textbox(label="Detailed Analysis Report"),  # Output: text report
-        gr.HTML(label="Download Report"),  # Output: downloadable link
-        gr.Textbox(label="Email Content (To Be Sent to Doctor)"),  # Output: email content
-        gr.HTML(label="WhatsApp Link")  # Output: WhatsApp link
+        gr.Image(label="Annotated Image"),
+        gr.Textbox(label="Detailed Analysis Report"),
+        gr.HTML(label="Download Report"),
+        gr.Textbox(label="Email Content (To Be Sent to Doctor)"),
+        gr.HTML(label="WhatsApp Link")
     ],
-    title="Remote Eye Image Analysis for Patients in India (Supports Poor-Quality Images)",
-    description="Upload an eye image (e.g., fundus image or diagram) to analyze anatomical features. The system supports poor-quality images with enhanced preprocessing. The report will be prepared for the doctor to review, with options to send via email or WhatsApp.",
-    live=True  # Enable live updates for real-time analysis
+    title="EyeScanIndia: Remote Eye Image Analysis",
+    description="""
+    Upload an eye image (e.g., fundus image or diagram) to analyze anatomical features. 
+    Supports poor-quality images with enhanced preprocessing. 
+    Enter patient name, doctor's email, and WhatsApp number to generate a report.
+    **Note**: Ensure compliance with India’s DPDP Act for medical data. 
+    For best results, use high-quality fundus images.
+    """,
+    allow_flagging="never"
 )
 
-# Launch the Gradio interface
-interface.launch()
+# Launch the app for Hugging Face
+if __name__ == "__main__":
+    interface.launch(server_name="0.0.0.0", server_port=7860)