app.py

from fastai.vision.all import *
import gradio as gr
from PIL import Image
import pydicom
import numpy as np

path = "my_exported_model.pkl"

def process_dicom(fn):
    # Read DICOM file
    dcm = pydicom.dcmread(fn)

    # Access pixel data and scale it to 8-bit (0-255) range
    pixel_array = dcm.pixel_array
    pixel_array = pixel_array - np.min(pixel_array)
    pixel_array = pixel_array / np.max(pixel_array)
    pixel_array = (pixel_array * 255).astype(np.uint8)

    # Convert to PIL Image for further processing
    img = Image.fromarray(pixel_array).convert('L')

    # Resize the image (example: resize to 224x224)
    img = img.resize((512, 512))
    return PILImage(img)


def get_x(x):
    return process_dicom(images_path/f"{x['image_id']}")

def get_y(x):
    return x["class_name"]

# Function to process PNG images, mimicking your training preprocessing
def process_png(img):
    # If the image is a NumPy array, convert it to a PIL Image
    if isinstance(img, np.ndarray):
        img = Image.fromarray(img)

    # Convert to grayscale
    img = img.convert('L')

    # Resize the image to 512x512
    img = img.resize((512, 512))
    return PILImage(img)


# Prediction function
def predict_image(image):
    # Process the input image
    processed_img = process_png(image)

    learn = load_learner(path)

    # Get predictions
    pred, idx, probs = learn.predict(processed_img)

    categories = learn.dls.vocab
    return dict(zip(categories, map(float, probs)))


description = "This model achieved 0.906 accuracy on a test set of 600 images."

iface = gr.Interface(fn=predict_image, inputs="image", outputs=gr.Label(), 
                     description=description)

iface.launch()