Initial commit

Dockerfile

@@ -0,0 +1,20 @@
+# Use an official Python runtime
+FROM python:3.10
+
+# Set the working directory
+WORKDIR /app
+
+# Copy the requirements file
+COPY requirements.txt .
+
+# Install dependencies
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the FastAPI app file
+COPY app.py .
+
+# Expose the port FastAPI runs on
+EXPOSE 7860
+
+# Command to run FastAPI using Uvicorn
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,81 @@
+from fastapi import FastAPI, File, UploadFile
+import torch
+from transformers import AutoProcessor, LlavaForConditionalGeneration
+from PIL import Image
+import io
+import base64
+
+# Initialize FastAPI app
+app = FastAPI()
+
+# Load the model and processor from Hugging Face
+model_name = "mervinpraison/Llama-3.2-11B-Vision-Radiology-mini"
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+processor = AutoProcessor.from_pretrained(model_name)
+model = LlavaForConditionalGeneration.from_pretrained(model_name).to(device)
+
+@app.post("/predict/")
+async def predict(file: UploadFile = File(...)):
+    try:
+        # Read image
+        image_bytes = await file.read()
+        image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+
+        # Convert image to base64 (for compatibility with reference implementation)
+        buffered = io.BytesIO()
+        image.save(buffered, format="JPEG")
+        base64_image = base64.b64encode(buffered.getvalue()).decode("utf-8")
+
+        # Step 1: Validate Image Type (Ensure it’s an X-ray or CT scan)
+        validation_prompt = "Is this a medical X-ray or CT scan? Answer only 'yes' or 'no'."
+        validation_inputs = processor(text=validation_prompt, images=image, return_tensors="pt").to(device)
+
+        with torch.no_grad():
+            validation_output = model.generate(**validation_inputs, max_new_tokens=10, temperature=0.1, top_p=0.7, top_k=50, repetition_penalty=1)
+
+        validation_result = processor.batch_decode(validation_output, skip_special_tokens=True)[0].strip().lower()
+
+        if "yes" not in validation_result:
+            return {"error": "Uploaded image is not an X-ray or CT scan. Please upload a valid medical imaging scan."}
+
+        # Step 2: Generate Structured Medical Analysis
+        analysis_prompt = """Please analyze this X-ray image and provide a detailed medical report using the following format:
+
+Type of X-ray:
+[Describe the type and orientation of the X-ray]
+
+Key Findings:
+• [List each finding on a new line with a bullet point]
+• [Focus on normal and abnormal findings]
+• [Include major anatomical structures]
+
+Potential Conditions:
+• [List potential conditions based on findings]
+• [Include likelihood assessments]
+
+Recommendations:
+• [Provide any follow-up recommendations]
+
+Please provide the analysis in plain text without any special characters or markdown formatting."""
+
+        analysis_inputs = processor(text=analysis_prompt, images=image, return_tensors="pt").to(device)
+
+        with torch.no_grad():
+            analysis_output = model.generate(**analysis_inputs, max_new_tokens=512, temperature=0.7, top_p=0.7, top_k=50, repetition_penalty=1)
+
+        analysis_content = processor.batch_decode(analysis_output, skip_special_tokens=True)[0]
+
+        # Step 3: Clean Up Response (Remove special characters, markdown formatting)
+        cleaned_analysis = (
+            analysis_content.replace("**", "")  # Remove double asterisks
+            .replace("*", "•")  # Replace single asterisks with bullet points
+            .replace("_", "")  # Remove underscores
+            .replace("#", "")  # Remove markdown headers
+            .strip()
+        )
+
+        return {"analysis": cleaned_analysis}
+
+    except Exception as e:
+        return {"error": str(e)}

requirement.txt

@@ -0,0 +1,6 @@
+fastapi
+uvicorn
+torch
+transformers
+pillow
+python-multipart