dicm endpoint

main.py

@@ -2,6 +2,82 @@ from fastapi import FastAPI, File, UploadFile
 from fastapi.responses import JSONResponse
 from inference_service import generate_report_serviceFn
 from PIL import Image
+import pydicom
+import tempfile
+from fastapi import HTTPException
+import numpy as np
+
+
+def is_xray_dicom(dicom_data):
+    """
+    Check if the DICOM file is an X-ray image based on metadata
+    """
+    try:
+        # Check modality
+        if hasattr(dicom_data, 'Modality') and dicom_data.Modality in ['CR', 'DX', 'RF']:
+            return True
+        
+        # Check body part examined for chest X-ray indicators
+        if hasattr(dicom_data, 'BodyPartExamined'):
+            chest_keywords = ['CHEST', 'THORAX', 'LUNG', 'PULMONARY']
+            body_part = str(dicom_data.BodyPartExamined).upper()
+            if any(keyword in body_part for keyword in chest_keywords):
+                return True
+        
+        # Check study description
+        if hasattr(dicom_data, 'StudyDescription'):
+            study_desc = str(dicom_data.StudyDescription).upper()
+            xray_keywords = ['X-RAY', 'XRAY', 'RADIOGRAPH', 'CHEST']
+            if any(keyword in study_desc for keyword in xray_keywords):
+                return True
+        
+        return False
+    except Exception:
+        return False
+
+def dicom_to_png(dicom_data):
+    """
+    Convert DICOM image data to PNG format
+    """
+    try:
+        # Ensure pixel data exists
+        if not hasattr(dicom_data, "PixelData"):
+            raise ValueError("No pixel data found in DICOM file.")
+        
+        # Get and squeeze pixel array
+        pixel_array = dicom_data.pixel_array
+        pixel_array = np.squeeze(pixel_array)  # Removes (1, 1, 512) → (512)
+
+        if pixel_array.ndim == 3:
+            pixel_array = pixel_array[pixel_array.shape[0] // 2]
+        if pixel_array.ndim == 1:
+            raise ValueError(f"Unsupported 1D image shape: {pixel_array.shape}")
+
+        # Normalize pixel values to 0-255 range
+        if pixel_array.dtype != np.uint8:
+            max_val = 2 ** dicom_data.get("BitsStored", 16) - 1
+            pixel_array = ((pixel_array - pixel_array.min()) / (max_val - pixel_array.min()) * 255).astype(np.uint8)
+
+        # Handle MONOCHROME1
+        if dicom_data.get("PhotometricInterpretation") == "MONOCHROME1":
+            pixel_array = 255 - pixel_array
+
+        # Convert to PIL image
+        if pixel_array.ndim == 2:
+            image = Image.fromarray(pixel_array, mode="L")
+        elif pixel_array.ndim == 3 and pixel_array.shape[-1] in [3, 4]:  # RGB or RGBA
+            image = Image.fromarray(pixel_array)
+        else:
+            raise ValueError(f"Unsupported image shape: {pixel_array.shape}")
+
+        return image.convert("RGB")
+
+    except Exception as e:
+        raise ValueError(f"Failed to convert DICOM to PNG: {str(e)}")
+
+
+
+
 app = FastAPI()
 @app.post("/generate_report")
 async def generate_report(file: UploadFile = File(...)):
@@ -16,3 +92,73 @@ async def generate_report(file: UploadFile = File(...)):
     except Exception as e:
         return JSONResponse({"error": str(e)}, status_code=500)
 
+
+@app.post("/generate_report_dicom")
+async def generate_report_dicom(file: UploadFile = File(...)):
+    """
+    Generate radiology report from DICOM file.
+    Checks if the DICOM is an X-ray, converts it to PNG, and generates a report.
+    """
+    try:
+        # Validate file type
+        if not file.filename.lower().endswith(('.dcm', '.dicom')):
+            raise HTTPException(status_code=400, detail="File must be a DICOM file (.dcm or .dicom)")
+        
+        # Read DICOM file
+        file_content = await file.read()
+        
+        # Create a temporary file to store DICOM data
+        with tempfile.NamedTemporaryFile(delete=False, suffix='.dcm') as temp_file:
+            temp_file.write(file_content)
+            temp_file_path = temp_file.name
+        
+        try:
+            # Read DICOM data
+            dicom_data = pydicom.dcmread(temp_file_path)
+            
+            # # Check if it's an X-ray
+            # if not is_xray_dicom(dicom_data):
+            #     return JSONResponse(
+            #         {"error": "DICOM file does not appear to be an X-ray image"},
+            #         status_code=400
+            #     )
+            
+            # Convert DICOM to PNG
+            image = dicom_to_png(dicom_data)
+            
+            # Generate the report using the service function
+            report = generate_report_serviceFn(image)
+            
+            # Get additional DICOM metadata for context
+            metadata = {}
+            if hasattr(dicom_data, 'PatientID'):
+                metadata['patient_id'] = str(dicom_data.PatientID)
+            if hasattr(dicom_data, 'StudyDate'):
+                metadata['study_date'] = str(dicom_data.StudyDate)
+            if hasattr(dicom_data, 'Modality'):
+                metadata['modality'] = str(dicom_data.Modality)
+            if hasattr(dicom_data, 'BodyPartExamined'):
+                metadata['body_part'] = str(dicom_data.BodyPartExamined)
+            
+            return JSONResponse({
+                "generated_report": report,
+                "dicom_metadata": metadata,
+                "image_info": {
+                    "width": image.width,
+                    "height": image.height,
+                    "mode": image.mode
+                }
+            })
+            
+        finally:
+            # Clean up temporary file
+            import os
+            try:
+                os.unlink(temp_file_path)
+            except:
+                pass
+                
+    except HTTPException:
+        raise
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=500)

requirements.txt

@@ -3,4 +3,9 @@ pillow
 torch
 transformers
 python-multipart
-uvicorn
+uvicorn
+nltk
+scikit-learn
+numpy
+pandas
+pydicom