Anupam007's picture
Create app.py
cba643c verified
import gradio as gr
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, models
from tensorflow.keras.applications import EfficientNetB0
import cv2
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from PIL import Image
import io
import base64
from datetime import datetime
import warnings
import json
from scipy import ndimage
from skimage import measure, morphology, filters
import plotly.graph_objects as go
import plotly.express as px
from plotly.subplots import make_subplots
import logging
import re
from typing import Dict, Tuple, Optional, List, Any
warnings.filterwarnings('ignore')
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Check GPU availability
print("GPU Available: ", tf.config.list_physical_devices('GPU'))
print("TensorFlow version:", tf.__version__)
# Constants
IMAGE_SIZE = 512
MIN_AGE = 0
MAX_AGE = 120
MAX_PATIENT_ID_LENGTH = 50
DEFAULT_CONFIDENCE_LEVEL = 0.95
Z_SCORE_95 = 1.96
Z_SCORE_99 = 2.58
NORMALIZATION_CLIP_MIN = -3
NORMALIZATION_CLIP_MAX = 3
CLAHE_CLIP_LIMIT = 3.0
CLAHE_TILE_GRID_SIZE = (16, 16)
# Clinical eye conditions with ICD-10 codes and severity levels
CLINICAL_CONDITIONS = {
'diabetic_retinopathy': {
'name': 'Diabetic Retinopathy',
'icd10': 'E11.31',
'severity_levels': ['Mild NPDR', 'Moderate NPDR', 'Severe NPDR', 'PDR'],
'urgency': 'high',
'description': 'Retinal vascular damage secondary to diabetes mellitus'
},
'diabetic_macular_edema': {
'name': 'Diabetic Macular Edema',
'icd10': 'E11.311',
'severity_levels': ['Mild', 'Moderate', 'Severe'],
'urgency': 'urgent',
'description': 'Macular thickening with retinal exudates secondary to diabetes'
},
'glaucoma': {
'name': 'Glaucoma',
'icd10': 'H40.9',
'severity_levels': ['Suspect', 'Early', 'Moderate', 'Advanced'],
'urgency': 'high',
'description': 'Progressive optic neuropathy with characteristic optic disc changes'
},
'age_related_macular_degeneration': {
'name': 'Age-Related Macular Degeneration',
'icd10': 'H35.30',
'severity_levels': ['Early', 'Intermediate', 'Advanced Dry', 'Wet AMD'],
'urgency': 'moderate',
'description': 'Progressive degeneration of the macula affecting central vision'
},
'macular_hole': {
'name': 'Macular Hole',
'icd10': 'H35.341',
'severity_levels': ['Stage 1', 'Stage 2', 'Stage 3', 'Stage 4'],
'urgency': 'urgent',
'description': 'Full-thickness defect in the neurosensory retina at the fovea'
},
'epiretinal_membrane': {
'name': 'Epiretinal Membrane',
'icd10': 'H35.37',
'severity_levels': ['Mild', 'Moderate', 'Severe'],
'urgency': 'moderate',
'description': 'Fibrocellular proliferation on the inner retinal surface'
},
'retinal_detachment': {
'name': 'Retinal Detachment',
'icd10': 'H33.9',
'severity_levels': ['Localized', 'Extensive', 'Total'],
'urgency': 'emergency',
'description': 'Separation of neurosensory retina from retinal pigment epithelium'
},
'retinal_vein_occlusion': {
'name': 'Retinal Vein Occlusion',
'icd10': 'H34.8',
'severity_levels': ['BRVO', 'CRVO', 'Ischemic', 'Non-ischemic'],
'urgency': 'urgent',
'description': 'Blockage of retinal venous circulation'
},
'posterior_uveitis': {
'name': 'Posterior Uveitis',
'icd10': 'H20.2',
'severity_levels': ['Mild', 'Moderate', 'Severe'],
'urgency': 'high',
'description': 'Inflammation of posterior uveal tract including choroid'
},
'normal': {
'name': 'Normal Fundus',
'icd10': 'Z01.00',
'severity_levels': ['Normal'],
'urgency': 'routine',
'description': 'No pathological findings detected'
}
}
class ClinicalRetinalAnalyzer:
def __init__(self, training_sample_size: Optional[int] = None):
"""
Initialize the clinical retinal analyzer.
Args:
training_sample_size: Size of training dataset for CI calculations
"""
self.model = self.create_clinical_model()
self.training_sample_size = training_sample_size
self.initialize_clinical_parameters()
def create_clinical_model(self):
"""Create an ensemble model for clinical accuracy"""
try:
# Primary model - EfficientNet for overall classification
base_model = EfficientNetB0(
weights='imagenet',
include_top=False,
input_shape=(IMAGE_SIZE, IMAGE_SIZE, 3)
)
base_model.trainable = False
# Unfreeze top layers for fine-tuning
for layer in base_model.layers[-20:]:
layer.trainable = True
model = models.Sequential([
base_model,
layers.GlobalAveragePooling2D(),
layers.BatchNormalization(),
layers.Dropout(0.4),
layers.Dense(
1024,
activation='relu',
kernel_regularizer=tf.keras.regularizers.l2(0.001)
),
layers.BatchNormalization(),
layers.Dropout(0.3),
layers.Dense(
512,
activation='relu',
kernel_regularizer=tf.keras.regularizers.l2(0.001)
),
layers.Dropout(0.2),
layers.Dense(
len(CLINICAL_CONDITIONS),
activation='sigmoid',
name='main_output'
)
])
# Compile with clinical-appropriate metrics
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
loss='binary_crossentropy',
metrics=['accuracy', 'precision', 'recall', 'auc']
)
return model
except Exception as e:
logger.error(f"Error creating model: {str(e)}")
return None
def initialize_clinical_parameters(self):
"""Initialize clinical thresholds and parameters"""
self.clinical_thresholds = {
'diabetic_retinopathy': 0.3,
'diabetic_macular_edema': 0.4,
'glaucoma': 0.35,
'age_related_macular_degeneration': 0.4,
'macular_hole': 0.5,
'epiretinal_membrane': 0.3,
'retinal_detachment': 0.6,
'retinal_vein_occlusion': 0.4,
'posterior_uveitis': 0.35,
'normal': 0.5
}
# Prevalence-based calibration factors
self.prevalence_factors = {
'diabetic_retinopathy': 0.85,
'diabetic_macular_edema': 0.90,
'glaucoma': 0.80,
'age_related_macular_degeneration': 0.75,
'macular_hole': 0.95,
'epiretinal_membrane': 0.80,
'retinal_detachment': 0.98,
'retinal_vein_occlusion': 0.85,
'posterior_uveitis': 0.85,
'normal': 0.70
}
# Sensitivity and specificity targets for clinical use
self.performance_targets = {
'sensitivity': 0.90, # High sensitivity for screening
'specificity': 0.85, # Good specificity to reduce false positives
'ppv': 0.80, # Positive predictive value
'npv': 0.95 # Negative predictive value
}
def validate_input_data(self, patient_id: str, patient_age: str) -> Tuple[str, int]:
"""
Validate and sanitize input data.
Args:
patient_id: Patient identifier
patient_age: Patient age as string
Returns:
Tuple of validated patient_id and patient_age
Raises:
ValueError: If validation fails
"""
# Validate Patient ID
if patient_id:
# Sanitize patient ID - remove special characters except alphanumeric,
# hyphens, and underscores
patient_id = re.sub(r'[^a-zA-Z0-9\-_]', '', patient_id)
patient_id = patient_id[:MAX_PATIENT_ID_LENGTH]
# Validate Patient Age
validated_age = None
if patient_age:
try:
validated_age = int(patient_age)
if validated_age < MIN_AGE or validated_age > MAX_AGE:
raise ValueError(
f"Patient age must be between {MIN_AGE} and {MAX_AGE}."
)
except (ValueError, TypeError):
raise ValueError("Invalid patient age. Must be a number.")
return patient_id, validated_age
def advanced_image_preprocessing(self, image) -> Tuple[
Optional[np.ndarray], float, str
]:
"""
Clinical-grade image preprocessing with quality assessment and error handling.
Args:
image: Input image (PIL Image or numpy array)
Returns:
Tuple of (processed_image, quality_score, quality_message)
"""
try:
# Convert to numpy array if PIL
if isinstance(image, Image.Image):
original_array = np.array(image)
else:
original_array = image
# Validate image
if len(original_array.shape) not in [2, 3]:
return None, 0.0, "Invalid image format: Must be RGB or grayscale"
# Ensure RGB format
if len(original_array.shape) == 2:
original_array = cv2.cvtColor(original_array, cv2.COLOR_GRAY2RGB)
# Image quality assessment
quality_score = self.assess_image_quality(original_array)
if quality_score < 0.5:
return (
None,
quality_score,
"Image quality insufficient for analysis (score < 0.5)"
)
# Resize to clinical standard
processed = cv2.resize(
original_array,
(IMAGE_SIZE, IMAGE_SIZE),
interpolation=cv2.INTER_LANCZOS4
)
logger.info(f"Resized image shape: {processed.shape}")
# Advanced preprocessing pipeline
if len(processed.shape) == 3:
# Green channel enhancement (best contrast for retinal features)
green_channel = processed[:, :, 1]
# Validate green channel
if green_channel.size == 0:
return None, quality_score, "Invalid green channel data"
# Apply CLAHE with clinical parameters
clahe = cv2.createCLAHE(
clipLimit=CLAHE_CLIP_LIMIT,
tileGridSize=CLAHE_TILE_GRID_SIZE
)
enhanced = clahe.apply(green_channel)
# Reconstruct RGB with enhanced green channel
processed[:, :, 1] = enhanced
# Vessel enhancement using morphological operations
processed = self.enhance_retinal_features(processed)
# Normalize with clinical standards
processed = processed.astype(np.float32)
# Use machine epsilon to prevent division by zero
std_val = np.std(processed)
epsilon = np.finfo(processed.dtype).eps
processed = (processed - np.mean(processed)) / (std_val + epsilon)
# Clip outliers
processed = np.clip(
processed,
NORMALIZATION_CLIP_MIN,
NORMALIZATION_CLIP_MAX
)
# Normalize to [0, 1]
processed = (processed + 3) / 6
return np.expand_dims(processed, axis=0), quality_score, "Quality acceptable"
except Exception as e:
logger.error(f"Error in image preprocessing: {str(e)}")
return None, 0.0, f"Error in image preprocessing: {str(e)}"
def assess_image_quality(self, image: np.ndarray) -> float:
"""
Assess image quality for clinical analysis.
Args:
image: Input image array
Returns:
Quality score between 0 and 1
"""
try:
if len(image.shape) == 3:
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
else:
gray = image
# Multiple quality metrics
metrics = {}
# 1. Sharpness (Laplacian variance)
metrics['sharpness'] = cv2.Laplacian(gray, cv2.CV_64F).var()
# 2. Contrast (RMS contrast)
metrics['contrast'] = gray.std()
# 3. Brightness distribution
metrics['brightness'] = np.mean(gray)
# 4. Dynamic range
metrics['dynamic_range'] = np.ptp(gray)
# Normalize and combine metrics
quality_score = min(1.0, (
min(metrics['sharpness'] / 500, 1.0) * 0.3 +
min(metrics['contrast'] / 50, 1.0) * 0.3 +
min(abs(metrics['brightness'] - 128) / 128, 1.0) * 0.2 +
min(metrics['dynamic_range'] / 255, 1.0) * 0.2
))
return quality_score
except Exception as e:
logger.error(f"Error assessing image quality: {str(e)}")
return 0.0
def enhance_retinal_features(self, image: np.ndarray) -> np.ndarray:
"""
Enhance retinal-specific features.
Args:
image: Input image array
Returns:
Enhanced image array
"""
try:
# Convert to LAB color space
lab = cv2.cvtColor(image, cv2.COLOR_RGB2LAB)
# Enhance L channel
l_channel = lab[:, :, 0]
# Apply bilateral filter to reduce noise while preserving edges
filtered = cv2.bilateralFilter(l_channel, 9, 75, 75)
# Enhance vessels using top-hat transform
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (15, 15))
tophat = cv2.morphologyEx(filtered, cv2.MORPH_TOPHAT, kernel)
enhanced = cv2.add(filtered, tophat)
lab[:, :, 0] = enhanced
# Convert back to RGB
enhanced_image = cv2.cvtColor(lab, cv2.COLOR_LAB2RGB)
return enhanced_image
except Exception as e:
logger.error(f"Error enhancing retinal features: {str(e)}")
return image
def clinical_prediction(self, processed_image: np.ndarray) -> Tuple[
Optional[Dict], str
]:
"""
Generate clinical predictions with confidence intervals.
Args:
processed_image: Preprocessed image array
Returns:
Tuple of (clinical_results, status_message)
"""
try:
if processed_image is None:
return None, "Processed image is None"
# Validate input shape
expected_shape = (1, IMAGE_SIZE, IMAGE_SIZE, 3)
if processed_image.shape != expected_shape:
return None, (
f"Invalid input shape: {processed_image.shape}, "
f"expected {expected_shape}"
)
# Check for invalid values
if np.any(np.isnan(processed_image)) or np.any(np.isinf(processed_image)):
return None, "Processed image contains NaN or infinite values"
# Check if model is initialized
if self.model is None:
return None, "Model not initialized"
# Get base predictions
logger.info("Running model prediction...")
predictions = self.model.predict(processed_image, verbose=0)[0]
logger.info(f"Predictions shape: {predictions.shape}, values: {predictions}")
# Apply clinical thresholds and generate refined predictions
clinical_results = {}
condition_keys = list(CLINICAL_CONDITIONS.keys())
if len(predictions) != len(condition_keys):
return None, (
f"Prediction length mismatch: {len(predictions)} "
f"vs {len(condition_keys)}"
)
for i, (condition_key, pred_value) in enumerate(
zip(condition_keys, predictions)
):
condition_info = CLINICAL_CONDITIONS[condition_key]
threshold = self.clinical_thresholds[condition_key]
# Calculate clinical probability with uncertainty
clinical_prob = self.apply_clinical_calibration(pred_value, condition_key)
# Determine severity if positive
severity = self.determine_severity(clinical_prob, condition_key)
clinical_results[condition_key] = {
'probability': float(clinical_prob),
'raw_score': float(pred_value),
'positive': clinical_prob >= threshold,
'severity': severity,
'confidence_interval': self.calculate_confidence_interval(
clinical_prob
),
'clinical_significance': self.assess_clinical_significance(
clinical_prob, condition_key
),
'condition_info': condition_info
}
return clinical_results, "Success"
except Exception as e:
logger.error(f"Error in clinical prediction: {str(e)}")
return None, f"Prediction failed: {str(e)}"
def apply_clinical_calibration(self, raw_prediction: float, condition_key: str) -> float:
"""
Apply clinical calibration based on real-world prevalence.
Args:
raw_prediction: Raw model prediction
condition_key: Condition identifier
Returns:
Calibrated probability
"""
try:
factor = self.prevalence_factors.get(condition_key, 0.80)
calibrated = raw_prediction * factor
return np.clip(calibrated, 0.0, 1.0)
except Exception as e:
logger.error(f"Error in clinical calibration: {str(e)}")
return 0.0
def determine_severity(self, probability: float, condition_key: str) -> str:
"""
Determine condition severity based on probability.
Args:
probability: Detection probability
condition_key: Condition identifier
Returns:
Severity level string
"""
try:
severity_levels = CLINICAL_CONDITIONS[condition_key]['severity_levels']
if probability < self.clinical_thresholds[condition_key]:
return 'Not detected'
elif probability < 0.5:
return severity_levels[0] if severity_levels else 'Mild'
elif probability < 0.7:
return severity_levels[1] if len(severity_levels) > 1 else 'Moderate'
elif probability < 0.85:
return severity_levels[2] if len(severity_levels) > 2 else 'Severe'
else:
return severity_levels[-1] if severity_levels else 'Severe'
except Exception as e:
logger.error(f"Error determining severity: {str(e)}")
return 'N/A'
def calculate_confidence_interval(
self,
probability: float,
confidence_level: float = DEFAULT_CONFIDENCE_LEVEL
) -> Dict[str, float]:
"""
Calculate confidence interval for predictions.
Args:
probability: Detection probability
confidence_level: Confidence level (default 0.95)
Returns:
Dictionary with 'lower' and 'upper' bounds
"""
try:
# Check if training sample size is set
if self.training_sample_size is None:
logger.warning(
"Training sample size 'n' is not set. "
"Confidence intervals may be inaccurate."
)
return {'lower': 0.0, 'upper': 0.0}
# Wilson score interval calculation
n = self.training_sample_size
z = Z_SCORE_95 if confidence_level == 0.95 else Z_SCORE_99
p = probability
denominator = 1 + z**2/n
center = p + z**2/(2*n)
margin = z * np.sqrt(p*(1-p)/n + z**2/(4*n**2))
ci_lower = max(0, (center - margin) / denominator)
ci_upper = min(1, (center + margin) / denominator)
return {'lower': ci_lower, 'upper': ci_upper}
except Exception as e:
logger.error(f"Error calculating confidence interval: {str(e)}")
return {'lower': 0.0, 'upper': 0.0}
def assess_clinical_significance(
self,
probability: float,
condition_key: str
) -> str:
"""
Assess clinical significance of findings.
Args:
probability: Detection probability
condition_key: Condition identifier
Returns:
Clinical significance assessment
"""
try:
condition_info = CLINICAL_CONDITIONS[condition_key]
urgency = condition_info['urgency']
if probability < self.clinical_thresholds[condition_key]:
return 'Not significant'
elif urgency == 'emergency' and probability > 0.7:
return 'Immediate referral required'
elif urgency == 'urgent' and probability > 0.6:
return 'Urgent referral recommended'
elif urgency == 'high' and probability > 0.5:
return 'Prompt evaluation needed'
else:
return 'Monitor and follow-up'
except Exception as e:
logger.error(f"Error assessing clinical significance: {str(e)}")
return 'Not significant'
# Initialize the clinical analyzer
# TODO: Set training_sample_size based on actual training data
analyzer = ClinicalRetinalAnalyzer(training_sample_size=None)
def generate_clinical_visualization(results: Dict) -> Tuple[
Optional[go.Figure], Optional[go.Figure]
]:
"""
Generate comprehensive clinical visualization with error handling.
Args:
results: Clinical analysis results
Returns:
Tuple of (probability_figure, confidence_figure)
"""
try:
if not results:
return None, None
# Extract data for visualization
conditions = []
probabilities = []
severities = []
urgencies = []
colors = []
for condition_key, result in results.items():
if result['positive'] or result['probability'] > 0.1:
conditions.append(CLINICAL_CONDITIONS[condition_key]['name'])
probabilities.append(result['probability'])
severities.append(result['severity'])
urgencies.append(CLINICAL_CONDITIONS[condition_key]['urgency'])
# Color coding by urgency
urgency_colors = {
'emergency': 'red',
'urgent': 'orange',
'high': 'yellow',
'moderate': 'lightblue',
'routine': 'green'
}
colors.append(
urgency_colors.get(
CLINICAL_CONDITIONS[condition_key]['urgency'],
'gray'
)
)
if not conditions:
conditions = ['Normal Fundus']
probabilities = [0.85]
colors = ['green']
# Create main probability chart
fig1 = go.Figure()
fig1.add_trace(go.Bar(
y=conditions,
x=probabilities,
orientation='h',
marker_color=colors,
text=[f'{p:.1%}' for p in probabilities],
textposition='auto',
name='Detection Probability'
))
fig1.update_layout(
title='Clinical Detection Probability',
xaxis_title='Probability',
yaxis_title='Conditions',
height=400,
margin=dict(l=200, r=50, t=50, b=50)
)
# Create confidence interval chart
fig2 = make_subplots(
rows=1, cols=2,
subplot_titles=('Confidence Intervals', 'Urgency Distribution'),
specs=[[{"secondary_y": False}, {"type": "pie"}]]
)
# Confidence intervals
for condition_key, result in results.items():
if result['positive']:
ci = result['confidence_interval']
condition_name = CLINICAL_CONDITIONS[condition_key]['name']
fig2.add_trace(
go.Scatter(
x=[ci['lower'], result['probability'], ci['upper']],
y=[condition_name, condition_name, condition_name],
mode='markers+lines',
name=condition_name,
line=dict(width=3),
marker=dict(size=[8, 12, 8])
),
row=1, col=1
)
# Urgency pie chart
urgency_counts = {}
for condition_key, result in results.items():
if result['positive']:
urgency = CLINICAL_CONDITIONS[condition_key]['urgency']
urgency_counts[urgency] = urgency_counts.get(urgency, 0) + 1
if urgency_counts:
urgency_colors_pie = {
'emergency': 'red',
'urgent': 'orange',
'high': 'yellow',
'moderate': 'lightblue',
'routine': 'green'
}
pie_colors = [urgency_colors_pie.get(k, 'gray') for k in urgency_counts.keys()]
fig2.add_trace(
go.Pie(
labels=list(urgency_counts.keys()),
values=list(urgency_counts.values()),
marker_colors=pie_colors
),
row=1, col=2
)
else:
# Fallback for no positive findings
fig2.add_trace(
go.Pie(
labels=['Normal'],
values=[1],
marker_colors=['green']
),
row=1, col=2
)
fig2.update_layout(height=400, showlegend=True)
return fig1, fig2
except Exception as e:
logger.error(f"Error in visualization: {str(e)}")
return None, None
def generate_clinical_report(
results: Dict,
image_quality: float,
patient_info: Optional[Dict] = None
) -> str:
"""
Generate comprehensive clinical report.
Args:
results: Clinical analysis results
image_quality: Image quality score
patient_info: Optional patient information
Returns:
Formatted clinical report string
"""
try:
if not results:
return "Error: Unable to generate clinical report."
# Count positive findings
positive_findings = [k for k, v in results.items() if v['positive']]
high_priority = [
k for k in positive_findings
if CLINICAL_CONDITIONS[k]['urgency'] in ['emergency', 'urgent']
]
report = f"""
# CLINICAL RETINAL ANALYSIS REPORT
## EXAMINATION DETAILS
- **Date & Time:** {datetime.now().strftime('%Y-%m-%d %H:%M:%S UTC')}
- **Analysis System:** AI-Assisted Retinal Screening v2.0
- **Image Quality Score:** {image_quality:.2f}/1.00 ({'Acceptable' if image_quality > 0.5 else 'Suboptimal'})
- **Analysis Method:** Deep Learning Ensemble (EfficientNet + Clinical Calibration)
"""
if patient_info:
report += f"""## PATIENT INFORMATION
- **Patient ID:** {patient_info.get('id', 'Not provided')}
- **Age:** {patient_info.get('age', 'Not provided')}
- **Medical History:** {patient_info.get('history', 'Not provided')}
"""
# Executive Summary
report += "## EXECUTIVE SUMMARY\n\n"
if high_priority:
report += "🚨 **URGENT FINDINGS DETECTED**\n\n"
for condition_key in high_priority:
condition_info = CLINICAL_CONDITIONS[condition_key]
result = results[condition_key]
ci = result['confidence_interval']
report += f"- **{condition_info['name']}** (ICD-10: {condition_info['icd10']})\n"
report += f" - Probability: {result['probability']:.1%} (CI: {ci['lower']:.1%}-{ci['upper']:.1%})\n"
report += f" - Severity: {result['severity']}\n"
report += f" - Action: {result['clinical_significance']}\n"
report += f" - Description: {condition_info['description']}\n\n"
else:
report += "✅ **No urgent findings detected**\n\n"
if positive_findings:
report += "Non-urgent findings detected requiring monitoring or follow-up.\n\n"
else:
report += "No pathological findings detected. Routine follow-up recommended.\n\n"
# Detailed Findings
report += "## DETAILED CLINICAL FINDINGS\n\n"
for condition_key, result in results.items():
condition_info = CLINICAL_CONDITIONS[condition_key]
ci = result['confidence_interval']
report += f"### {condition_info['name']} (ICD-10: {condition_info['icd10']})\n"
report += f"- **Detection Status:** {'Positive' if result['positive'] else 'Negative'}\n"
report += f"- **Probability:** {result['probability']:.1%} (95% CI: {ci['lower']:.1%}-{ci['upper']:.1%})\n"
report += f"- **Severity:** {result['severity']}\n"
report += f"- **Clinical Significance:** {result['clinical_significance']}\n"
report += f"- **Description:** {condition_info['description']}\n"
report += f"- **Urgency Level:** {condition_info['urgency'].capitalize()}\n\n"
# Recommendations
report += "## CLINICAL RECOMMENDATIONS\n\n"
if high_priority:
report += "- **Immediate Action:** Urgent referral to retina specialist recommended.\n"
report += "- **Diagnostic Confirmation:** Confirm findings with clinical examination and additional imaging (OCT, FFA if indicated).\n"
else:
report += "- **Follow-up:** Routine ophthalmologic examination recommended based on clinical guidelines.\n"
report += "- **Monitoring:** Regular screening as per patient risk factors and age.\n"
report += f"- **Image Quality Note:** Ensure high-quality fundus photography for optimal analysis (current quality: {image_quality:.2f}).\n"
# Performance Metrics
report += "\n## SYSTEM PERFORMANCE METRICS\n"
report += f"- **Sensitivity Target:** {analyzer.performance_targets['sensitivity']*100:.0f}%\n"
report += f"- **Specificity Target:** {analyzer.performance_targets['specificity']*100:.0f}%\n"
report += f"- **Positive Predictive Value Target:** {analyzer.performance_targets['ppv']*100:.0f}%\n"
report += f"- **Negative Predictive Value Target:** {analyzer.performance_targets['npv']*100:.0f}%\n"
report += "\n**Note:** This report is generated by an AI-assisted system and must be reviewed by a qualified ophthalmologist. Results are intended for clinical decision support and not as a definitive diagnosis."
return report
except Exception as e:
logger.error(f"Error generating clinical report: {str(e)}")
return f"Error: Unable to generate clinical report due to {str(e)}"
def analyze_retinal_image(
image_input: Any,
patient_id: str = "",
patient_age: str = "",
medical_history: str = ""
) -> Tuple[str, Optional[go.Figure], Optional[go.Figure]]:
"""
Main function to analyze retinal image and generate clinical output.
Args:
image_input: Input image (PIL Image, numpy array, or file path)
patient_id: Patient identifier
patient_age: Patient age as string
medical_history: Patient medical history
Returns:
Tuple of (clinical_report, probability_figure, confidence_figure)
"""
try:
# Validate patient inputs
validated_id, validated_age = analyzer.validate_input_data(patient_id, patient_age)
patient_info = {
'id': validated_id or 'Not provided',
'age': validated_age or 'Not provided',
'history': medical_history or 'Not provided'
}
# Preprocess image
processed_image, quality_score, quality_message = analyzer.advanced_image_preprocessing(image_input)
if processed_image is None:
return (
f"Error: Image preprocessing failed. {quality_message}",
None,
None
)
# Perform clinical prediction
results, status = analyzer.clinical_prediction(processed_image)
if results is None:
return (
f"Error: Analysis failed. {status}",
None,
None
)
# Generate visualizations
prob_fig, conf_fig = generate_clinical_visualization(results)
# Generate clinical report
report = generate_clinical_report(results, quality_score, patient_info)
return report, prob_fig, conf_fig
except Exception as e:
logger.error(f"Error in retinal image analysis: {str(e)}")
return (
f"Error: Analysis failed due to {str(e)}",
None,
None
)
def create_gradio_interface():
"""
Create Gradio interface for clinical use.
Returns:
Gradio interface object
"""
with gr.Blocks(title="Clinical Retinal Analysis System") as interface:
gr.Markdown(
"""
# Clinical Retinal Analysis System
AI-assisted retinal screening for medical professionals. Upload a fundus image and enter patient details for comprehensive analysis.
"""
)
with gr.Row():
with gr.Column(scale=2):
image_input = gr.Image(type="pil", label="Upload Fundus Image")
patient_id = gr.Textbox(label="Patient ID")
patient_age = gr.Textbox(label="Patient Age")
medical_history = gr.Textbox(label="Medical History", lines=3)
analyze_button = gr.Button("Analyze Retinal Image")
with gr.Column(scale=3):
report_output = gr.Markdown(label="Clinical Report")
prob_plot = gr.Plot(label="Detection Probabilities")
conf_plot = gr.Plot(label="Confidence Intervals & Urgency")
analyze_button.click(
fn=analyze_retinal_image,
inputs=[image_input, patient_id, patient_age, medical_history],
outputs=[report_output, prob_plot, conf_plot]
)
return interface
# Launch the interface
if __name__ == "__main__":
interface = create_gradio_interface()
interface.launch()