app.py

import gradio as gr
import numpy as np
from PIL import Image, ImageEnhance
import cv2
import torch
import os
import requests
from tqdm import tqdm
from segment_anything import sam_model_registry, SamPredictor
import matplotlib.pyplot as plt

# Define o CSS customizado
css = """
.gradio-container {
    font-family: 'Inter', sans-serif;
    max-width: 1200px !important;
    margin: auto;
}

.main-container {
    background-color: #ffffff;
    border-radius: 15px;
    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
    padding: 20px;
    margin: 20px;
}

.header {
    background: linear-gradient(90deg, #2563eb 0%, #3b82f6 100%);
    color: white;
    padding: 20px;
    border-radius: 10px;
    margin-bottom: 20px;
    text-align: center;
}

.controls {
    background-color: #f8fafc;
    padding: 15px;
    border-radius: 10px;
    margin: 10px 0;
}

.slider-label {
    color: #374151;
    font-weight: 500;
}

.button-primary {
    background: linear-gradient(90deg, #2563eb 0%, #3b82f6 100%);
    border: none;
    padding: 10px 20px;
    border-radius: 8px;
    color: white;
    font-weight: 600;
    cursor: pointer;
    transition: transform 0.2s;
}

.button-primary:hover {
    transform: translateY(-2px);
}

.output-container {
    background-color: #f8fafc;
    padding: 15px;
    border-radius: 10px;
    margin-top: 20px;
}

.analysis-text {
    font-family: 'Mono', monospace;
    background-color: #1e293b;
    color: #f8fafc;
    padding: 15px;
    border-radius: 8px;
    white-space: pre-wrap;
}

.image-input, .image-output {
    border-radius: 10px;
    overflow: hidden;
    border: 2px solid #e2e8f0;
}
"""

def download_sam_model():
    """Download SAM model if not already present"""
    model_path = "sam_vit_h_4b8939.pth"
    if not os.path.exists(model_path):
        print("Baixando modelo SAM... Isso pode levar alguns minutos.")
        url = "https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth"
        response = requests.get(url, stream=True)
        total_size = int(response.headers.get('content-length', 0))
        
        with open(model_path, 'wb') as file, tqdm(
            desc=model_path,
            total=total_size,
            unit='iB',
            unit_scale=True,
            unit_divisor=1024,
        ) as progress_bar:
            for data in response.iter_content(chunk_size=1024):
                size = file.write(data)
                progress_bar.update(size)
    return model_path

def load_sam_model():
    """Load SAM model with automatic download"""
    DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    MODEL_TYPE = "vit_h"
    
    checkpoint_path = download_sam_model()
    
    print(f"Carregando modelo SAM no dispositivo: {DEVICE}")
    sam = sam_model_registry[MODEL_TYPE](checkpoint=checkpoint_path)
    sam.to(device=DEVICE)
    predictor = SamPredictor(sam)
    return predictor

def enhance_xray(image, brightness=1.0, contrast=1.0, sharpness=1.0):
    """Enhance X-ray image with adjustable parameters"""
    if len(image.shape) == 2:
        image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
    
    img = Image.fromarray(image)
    
    # Apply enhancements
    img = ImageEnhance.Brightness(img).enhance(brightness)
    img = ImageEnhance.Contrast(img).enhance(contrast)
    img = ImageEnhance.Sharpness(img).enhance(sharpness)
    
    return np.array(img)

def generate_dense_points(image, num_points=100):
    """Gera uma grade densa de pontos para melhor segmentação"""
    h, w = image.shape[:2]
    points = []
    
    # Grade regular
    num_x = int(np.sqrt(num_points * w / h))
    num_y = int(num_points / num_x)
    
    x_points = np.linspace(w * 0.05, w * 0.95, num_x)
    y_points = np.linspace(h * 0.05, h * 0.95, num_y)
    
    # Grade regular
    for x in x_points:
        for y in y_points:
            points.append([x, y])
    
    # Adiciona pontos aleatórios para melhor cobertura
    num_random = num_points - len(points)
    if num_random > 0:
        random_x = np.random.uniform(w * 0.05, w * 0.95, num_random)
        random_y = np.random.uniform(h * 0.05, h * 0.95, num_random)
        for x, y in zip(random_x, random_y):
            points.append([x, y])
    
    return np.array(points)

def show_mask(mask, image, alpha=0.5):
    """Aplica uma máscara colorida sobre a imagem com transparência ajustável"""
    # Gera uma cor aleatória vibrante
    hue = np.random.random()
    color = plt.cm.hsv(hue)[:3]
    color = np.array([*color, alpha])
    
    h, w = mask.shape[-2:]
    mask_image = mask.reshape(h, w, 1) * color.reshape(1, 1, -1)
    
    if image.shape[-1] == 3:
        img_rgba = cv2.cvtColor(image, cv2.COLOR_RGB2RGBA)
    else:
        img_rgba = image
        
    mask_image_uint8 = (mask_image * 255).astype(np.uint8)
    
    # Mistura a máscara com a imagem usando alpha blending
    bg = img_rgba.astype(float)
    fg = mask_image_uint8.astype(float)
    
    alpha = fg[..., 3:] / 255.0
    fg_scaled = fg[..., :3] * alpha
    bg_scaled = bg[..., :3] * (1 - alpha)
    
    result = (fg_scaled + bg_scaled).astype(np.uint8)
    return cv2.cvtColor(result, cv2.COLOR_RGBA2RGB)

def segment_teeth(image, predictor):
    """Segment teeth using SAM2 with improved detection"""
    if len(image.shape) == 2:
        image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
    
    predictor.set_image(image)
    
    # Gerar pontos densos para melhor cobertura
    input_points = generate_dense_points(image, num_points=150)
    input_labels = np.ones(len(input_points))
    
    # Parâmetros ajustados para melhor detecção
    masks, scores, logits = predictor.predict(
        point_coords=input_points,
        point_labels=input_labels,
        multimask_output=True
    )
    
    # Filtrar máscaras por score
    filtered_masks = []
    filtered_scores = []
    for mask, score in zip(masks, scores):
        if score > 0.5:  # Ajuste este limiar conforme necessário
            filtered_masks.append(mask)
            filtered_scores.append(score)
    
    # Criar visualização da segmentação
    final_image = image.copy()
    
    # Ordenar máscaras por tamanho para melhor visualização
    mask_sizes = [np.sum(mask) for mask in filtered_masks]
    sorted_indices = np.argsort(mask_sizes)[::-1]
    
    for idx in sorted_indices:
        final_image = show_mask(filtered_masks[idx], final_image, alpha=0.4)
    
    # Desenhar pontos de referência menores e mais discretos
    for point in input_points:
        cv2.circle(final_image, (int(point[0]), int(point[1])), 2, (0, 127, 255), -1)
    
    return final_image, filtered_masks

def analyze_xray(image, brightness, contrast, sharpness, enable_segmentation):
    """Main function to process X-ray images"""
    if image is None:
        return None, "Por favor, carregue uma imagem."
    
    try:
        # Enhance image
        enhanced = enhance_xray(image, brightness, contrast, sharpness)
        
        # Initialize result
        result = enhanced
        
        # Perform segmentation if enabled
        if enable_segmentation:
            # Ensure predictor is initialized
            global sam_predictor
            if 'sam_predictor' not in globals():
                sam_predictor = load_sam_model()
            
            result, masks = segment_teeth(enhanced, sam_predictor)
            seg_status = "Segmentação realizada com sucesso"
            num_segments = len(masks)
            
            # Calcular área total segmentada
            total_area = sum(mask.sum() for mask in masks)
            image_area = image.shape[0] * image.shape[1]
            coverage_percent = (total_area / image_area) * 100
            
        else:
            seg_status = "Segmentação desativada"
            num_segments = 0
            coverage_percent = 0
        
        # Generate analysis text
        analysis = f"""📊 Análise da Radiografia:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
�밝 Parâmetros de Imagem:
   • Brilho: {brightness:.1f}
   • Contraste: {contrast:.1f}
   • Nitidez: {sharpness:.1f}

🔬 Resultado da Segmentação:
   • Status: {seg_status}
   • Estruturas detectadas: {num_segments}
   • Cobertura da análise: {coverage_percent:.1f}%

📝 Observações:
   • Processamento concluído com sucesso
   • Pontos laranja indicam regiões de análise
   • Áreas coloridas mostram estruturas detectadas"""
        
        return result, analysis
        
    except Exception as e:
        import traceback
        error_trace = traceback.format_exc()
        print(error_trace)
        return None, f"Erro durante o processamento: {str(e)}"

# Create Gradio interface with modern design
with gr.Blocks(css=css, title="Analisador de Radiografias Dentárias") as app:
    with gr.Column(elem_classes="main-container"):
        # Header
        with gr.Column(elem_classes="header"):
            gr.Markdown("# 🦷 Analisador Inteligente de Radiografias Dentárias")
            gr.Markdown("""
            Análise avançada de radiografias com Inteligência Artificial
            Utilizando o modelo SAM2 (Segment Anything Model 2)
            """)
        
        with gr.Row():
            # Input Column
            with gr.Column(scale=1):
                with gr.Group(elem_classes="controls"):
                    input_image = gr.Image(
                        label="Radiografia Original",
                        type="numpy",
                        elem_classes="image-input"
                    )
                    with gr.Column():
                        brightness = gr.Slider(
                            minimum=0.1, maximum=2.0, value=1.0,
                            label="Brilho",
                            elem_classes="slider-label"
                        )
                        contrast = gr.Slider(
                            minimum=0.1, maximum=2.0, value=1.0,
                            label="Contraste",
                            elem_classes="slider-label"
                        )
                        sharpness = gr.Slider(
                            minimum=0.1, maximum=2.0, value=1.0,
                            label="Nitidez",
                            elem_classes="slider-label"
                        )
                        enable_segmentation = gr.Checkbox(
                            label="Ativar Segmentação Inteligente",
                            value=True
                        )
                    analyze_btn = gr.Button(
                        "🔍 Analisar Radiografia",
                        elem_classes="button-primary"
                    )
            
            # Output Column
            with gr.Column(scale=1):
                with gr.Group(elem_classes="output-container"):
                    output_image = gr.Image(
                        label="Resultado da Análise",
                        type="numpy",
                        elem_classes="image-output"
                    )
                    analysis_text = gr.Textbox(
                        label="Análise Detalhada",
                        lines=8,
                        elem_classes="analysis-text"
                    )
        
        # Footer
        gr.Markdown("""
        ### 📋 Instruções de Uso
        1. Carregue uma radiografia dental
        2. Ajuste os parâmetros de imagem conforme necessário
        3. Ative a segmentação inteligente para detectar estruturas
        4. Clique em "Analisar Radiografia"
        
        ⚠️ Nota: O primeiro uso pode levar alguns minutos para baixar o modelo.
        """)
    
    analyze_btn.click(
        analyze_xray,
        inputs=[input_image, brightness, contrast, sharpness, enable_segmentation],
        outputs=[output_image, analysis_text]
    )

# Launch the app
if __name__ == "__main__":
    sam_predictor = load_sam_model()
    app.launch()