Update app.py

app.py

@@ -3,19 +3,27 @@ import streamlit as st
 from PIL import Image
 import torch
 from torchvision import transforms, models
+import numpy as np
+from groq import Groq
 
-# Set up environment variable for Groq API
+# Set up environment variables
 os.environ["GROQ_API_KEY"] = "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"
 
+# Initialize Groq client
+client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
+
+# Load Pretrained Models
+@st.cache_resource
 # Load Pretrained Model for Organ Recognition
 @st.cache_resource
 def load_organ_model():
-    model = models.resnet18(pretrained=True)  # ResNet18 pretrained model
-    model.fc = torch.nn.Linear(model.fc.in_features, 4)  # Modify for 4 classes
-    model.eval()
+    model = models.resnet18(pretrained=True)  # Load pretrained ResNet18
+    num_features = model.fc.in_features       # Get the number of input features to the final layer
+    model.fc = torch.nn.Linear(num_features, 4)  # Modify the final layer for 4 classes
+    model.eval()  # Set the model to evaluation mode
     return model
 
-organ_model = load_organ_model()
+organ_model, chexnet_model = load_model()
 
 # Image Preprocessing
 def preprocess_image(image):
@@ -26,17 +34,49 @@ def preprocess_image(image):
     ])
     return transform(image).unsqueeze(0)
 
+# Groq API for AI Insights
+def get_ai_insights(text_prompt):
+    try:
+        response = client.chat.completions.create(
+            messages=[{"role": "user", "content": text_prompt}],
+            model="llama-3.3-70b-versatile"
+        )
+        return response.choices[0].message.content
+    except Exception as e:
+        return f"Error: {e}"
+        
 # Organ Recognition Prediction
 def predict_organ(image):
     with torch.no_grad():
         input_tensor = preprocess_image(image)
         output = organ_model(input_tensor)
-        classes = ["Lungs", "Heart", "Spine", "Other"]  # Example organ classes
+
+        # Check the output dimensions
+        st.write(f"Model output shape: {output.shape}")
+
+        # Ensure the output matches the number of classes
+        classes = ["Lungs", "Heart", "Spine", "Other"]
+        if output.size(1) != len(classes):
+            raise ValueError(
+                f"Model output size ({output.size(1)}) does not match the number of classes ({len(classes)})."
+            )
+
+        # Get the prediction
+        prediction_index = output.argmax().item()
+        prediction = classes[prediction_index]
+        return prediction
+
+
+# Predict Normal/Abnormal
+def predict_normal_abnormal(image):
+    with torch.no_grad():
+        output = chexnet_model(preprocess_image(image))
+        classes = ["Normal", "Abnormal"]
         prediction = classes[output.argmax().item()]
     return prediction
 
 # Streamlit App
-st.title("X-ray Organ Recognition App")
+st.title("Medical X-ray Analysis App")
 st.sidebar.title("Navigation")
 task = st.sidebar.radio("Select a task", ["Upload X-ray", "AI Insights"])
 
@@ -48,12 +88,24 @@ if task == "Upload X-ray":
         st.image(image, caption="Uploaded X-ray", use_column_width=True)
 
         # Predict Organ
-        st.subheader("Step 1: Identify the Organ in the X-ray")
+        st.subheader("Step 1: Identify the Organ")
         organ = predict_organ(image)
         st.write(f"Predicted Organ: **{organ}**")
 
+        # Predict Normal/Abnormal
+        st.subheader("Step 2: Analyze the X-ray")
+        classification = predict_normal_abnormal(image)
+        st.write(f"X-ray Status: **{classification}**")
+
+        if classification == "Abnormal":
+            st.subheader("Step 3: AI-Based Insights")
+            ai_prompt = f"Explain why this X-ray of the {organ} is abnormal."
+            insights = get_ai_insights(ai_prompt)
+            st.write(insights)
+
 elif task == "AI Insights":
     st.subheader("Ask AI")
     user_input = st.text_area("Enter your query for AI insights")
     if user_input:
-        st.write("AI insights will be generated here.")  # Placeholder for AI logic
+        response = get_ai_insights(user_input)
+        st.write(response)