Add pneumonia detection app with Grad-CAM

app.py

@@ -4,9 +4,12 @@ import torchvision.transforms as T
 import torch.nn as nn
 import torch.nn.functional as F
 import gradio as gr
-from pytorch_grad_cam import GradCAM
-from pytorch_grad_cam.utils.image import show_cam_on_image
+# Grad-CAM imports removed for simplified UI
 from PIL import Image
+import requests
+import os
+import base64
+import io
 
 
 # Define CNN
@@ -53,35 +56,193 @@ transform = T.Compose([T.Resize((224,224)),
                        T.ToTensor(),
                        T.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5])])
 
-# Upload and visualize an image
-def predict_gradcam(image):
-    # prediction
-    img  = image.convert("RGB")
-    plt.imshow(image); plt.axis('off'); plt.show()
-    tensor = transform(img).unsqueeze(0).to(device)
+# Simplified prediction function without Grad-CAM
+def predict_pneumonia(image):
+    # Convert image to RGB
+    img = image.convert("RGB")
+    
+    # HERE IS WHERE THE IMAGE ENTERS THE MODEL:
+    # 1. Apply transforms (resize to 224x224, normalize)
+    tensor = transform(img).unsqueeze(0).to(device)  # Shape: [1, 3, 224, 224]
+    
+    # 2. Pass through the model
     with torch.no_grad():
         p = torch.softmax(model(tensor), dim=1)[0,1].item()
-    prob= f"{p:.3f}"
-    label= f"{'PNEUMONIA' if p>0.5 else 'NORMAL'}"
-
-    # Grad-CAM
-    target_layer = model.conv2
-    input_tensor = transform(img).unsqueeze(0).to(device)
-    cam = GradCAM(model=model, target_layers=[target_layer])
-    grayscale_cam = cam(input_tensor=input_tensor)[0]
-    img_np = np.array(img.resize((224,224)), dtype=np.float32)/255.0
-    heatmap = show_cam_on_image(img_np, grayscale_cam, use_rgb=True)
-    heatmap_pil = Image.fromarray(heatmap)
-
-    return prob, label, heatmap_pil
-
-demo = gr.Interface(
-    fn=predict_gradcam,
+    
+    # Format results
+    prob = f"{p:.3f}"
+    label = f"{'PNEUMONIA' if p>0.5 else 'NORMAL'}"
+    confidence = f"{p*100:.1f}%" if p > 0.5 else f"{(1-p)*100:.1f}%"
+
+    return prob, label, confidence
+
+# MedGemma Chatbot functionality
+def image_to_base64(image):
+    """Convert PIL image to base64 string"""
+    buffer = io.BytesIO()
+    image.save(buffer, format="JPEG")
+    img_bytes = buffer.getvalue()
+    img_base64 = base64.b64encode(img_bytes).decode()
+    return f"data:image/jpeg;base64,{img_base64}"
+
+def query_medgemma(message, history, image=None):
+    """Query MedGemma endpoint with text and optional image"""
+    
+    # Your endpoint URL
+    endpoint_url = "https://t911ok4t5x994zcu.us-east-1.aws.endpoints.huggingface.cloud"
+    
+    # Headers with your HF token
+    headers = {
+        "Authorization": f"Bearer {os.getenv('HUGGINGFACE_TOKEN')}",
+        "Content-Type": "application/json"
+    }
+    
+    # Prepare the message content
+    content = []
+    
+    # Add image if provided
+    if image is not None:
+        image_base64 = image_to_base64(image)
+        content.append({
+            "type": "image_url",
+            "image_url": {"url": image_base64}
+        })
+    
+    # Add text message
+    content.append({
+        "type": "text", 
+        "text": message
+    })
+    
+    # Prepare payload
+    payload = {
+        "model": "tgi",
+        "messages": [
+            {
+                "role": "user",
+                "content": content
+            }
+        ],
+        "max_tokens": 500,
+        "temperature": 0.7
+    }
+    
+    try:
+        response = requests.post(endpoint_url, headers=headers, json=payload, timeout=30)
+        
+        if response.status_code == 200:
+            result = response.json()
+            if "choices" in result and len(result["choices"]) > 0:
+                return result["choices"][0]["message"]["content"]
+            else:
+                return "Lo siento, no pude obtener una respuesta del modelo."
+        else:
+            return f"Error del endpoint: {response.status_code}. El modelo puede estar escalado a cero - intenta de nuevo en unos segundos."
+            
+    except requests.exceptions.Timeout:
+        return "Timeout: El modelo está despertando, intenta de nuevo en unos segundos."
+    except Exception as e:
+        return f"Error de conexión: {str(e)}"
+
+def medical_chat(message, history, uploaded_image):
+    """Handle medical chat with context from pneumonia detection"""
+    
+    # Add context about pneumonia detection if there's an image
+    context_message = message
+    if uploaded_image is not None:
+        context_message = f"""Como asistente médico especializado en radiología, analiza esta imagen de rayos X y responde: {message}
+        
+Contexto: Esta es una radiografía de tórax que puede mostrar signos de neumonía. Proporciona información médica precisa pero recuerda que siempre se debe consultar a un profesional médico."""
+    
+    response = query_medgemma(context_message, history, uploaded_image)
+    
+    # Add the exchange to history
+    history.append([message, response])
+    return history, ""
+
+# Create the main pneumonia detection interface
+pneumonia_interface = gr.Interface(
+    fn=predict_pneumonia,
     inputs=gr.Image(type="pil", label="Upload Chest X-ray"),
-    outputs=[gr.Textbox(label="Probability of Pneumonia"), gr.Label(label="Prediction"), gr.Image(label="Grad-CAM")],
+    outputs=[
+        gr.Textbox(label="Probability of Pneumonia"), 
+        gr.Label(label="Prediction"), 
+        gr.Textbox(label="Confidence")
+    ],
     title="🫁 Pneumonia Detection from Chest X-rays",
-    description="Upload a chest X-ray to see whether it shows signs of pneumonia. The model will predict the probability and show a Grad-CAM visualization of the most important regions.",
+    description="Upload a chest X-ray to detect signs of pneumonia using deep learning.",
     flagging_mode="never"
 )
 
+# Create the MedGemma chatbot interface
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🫁 RADOX - Sistema Inteligente de Detección de Neumonía")
+    gr.Markdown("### Análisis de Radiografías + Consulta Médica con IA")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            # Pneumonia Detection Section
+            gr.Markdown("## 🔍 Detección de Neumonía")
+            input_image = gr.Image(type="pil", label="Subir Radiografía de Tórax")
+            analyze_btn = gr.Button("🔬 Analizar Radiografía", variant="primary")
+            
+            with gr.Row():
+                prob_output = gr.Textbox(label="Probabilidad de Neumonía")
+                pred_output = gr.Label(label="Diagnóstico")
+                conf_output = gr.Textbox(label="Confianza")
+    
+    # Medical Chatbot Section
+    gr.Markdown("## 🤖 Consulta Médica con MedGemma")
+    gr.Markdown("*Haz preguntas sobre la radiografía o consultas médicas generales*")
+    
+    with gr.Row():
+        with gr.Column(scale=3):
+            chatbot = gr.Chatbot(
+                label="Chat Médico",
+                height=400,
+                show_label=True
+            )
+            
+            with gr.Row():
+                msg_input = gr.Textbox(
+                    label="Tu pregunta",
+                    placeholder="Ej: ¿Qué significan estos resultados? ¿Cuáles son los síntomas de neumonía?",
+                    scale=4
+                )
+                send_btn = gr.Button("Enviar", variant="primary", scale=1)
+        
+        with gr.Column(scale=1):
+            chat_image = gr.Image(
+                type="pil", 
+                label="Imagen para el chat (opcional)",
+                height=300
+            )
+            gr.Markdown("💡 **Tip:** Puedes subir la misma radiografía aquí para hacer preguntas específicas sobre ella.")
+    
+    # Event handlers
+    analyze_btn.click(
+        fn=predict_pneumonia,
+        inputs=[input_image],
+        outputs=[prob_output, pred_output, conf_output]
+    )
+    
+    send_btn.click(
+        fn=medical_chat,
+        inputs=[msg_input, chatbot, chat_image],
+        outputs=[chatbot, msg_input]
+    )
+    
+    msg_input.submit(
+        fn=medical_chat,
+        inputs=[msg_input, chatbot, chat_image],
+        outputs=[chatbot, msg_input]
+    )
+    
+    # Footer
+    gr.Markdown("""
+    ---
+    ⚠️ **Aviso Médico Importante**: Esta herramienta es solo para fines educativos y de apoyo diagnóstico. 
+    Siempre consulte con un profesional médico cualificado para obtener diagnósticos y tratamientos precisos.
+    """)
+
 demo.launch()

requirements.txt

@@ -1,8 +1,7 @@
-grad-cam==1.5.5
 gradio==4.44.1
 matplotlib==3.9.4
 numpy==2.0.2
-opencv-python==4.11.0.86
 pillow==10.4.0
 torch==2.7.1
-torchvision==0.22.1
+torchvision==0.22.1
+requests>=2.25.0