app.py

#ref: https://huggingface.co/spaces/Prgckwb/dicom-viewer/blob/main/app.py
#ref: https://huggingface.co/spaces/basilshaji/Lung_Nodule_Segmentation

import gradio as gr
import numpy as np
import polars as pl
import pydicom
from PIL import Image
from pydicom.errors import InvalidDicomError

import gradio as gr
import cv2
import requests
import os
import torch
import numpy as np
from yolov5.models.experimental import attempt_load
from yolov5.utils.general import non_max_suppression
from yolov5.utils.augmentations import letterbox

# Load YOLOv5 model (placeholder)
model_path = "best.pt"  # Path to your YOLOv5 model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # Use GPU if available
model = attempt_load(model_path, device=device)  # Placeholder for model loading
model.eval()  # Set the model to evaluation mode

def preprocess_image(image):
    img = letterbox(image, 640, stride=32, auto=True)[0]  # Resize and pad to 640x640
    img = img.transpose(2, 0, 1)[::-1]  # Convert BGR to RGB, 
    img = np.ascontiguousarray(img)
    img = torch.from_numpy(img).to(device)
    img = img.float()  # uint8 to fp16/32
    img /= 255.0  # 0 - 255 to 0.0 - 1.0
    if img.ndimension() == 3:
        img = img.unsqueeze(0)

    return img, image

def infer(model, img):
    with torch.no_grad():
        pred = model(img)[0]
    return pred

def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
    if ratio_pad is None:  # calculate from img0_shape
        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
    else:
        gain = ratio_pad[0]
        pad = ratio_pad[1]

    coords[:, [0, 2]] -= pad[0]  # x padding
    coords[:, [1, 3]] -= pad[1]  # y padding
    coords[:, :4] /= gain
    coords[:, :4].clip_(min=0, max=img1_shape[0])  # clip boxes
    return coords

def postprocess(pred, img0, img):    
    pred = non_max_suppression(pred, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False)
    results = []
    for det in pred:  # detections per image
        if len(det):
            det[:, :4] = scale_coords(img.shape[2:], det[:, :4], img0.shape).round()
            for *xyxy, conf, cls in reversed(det):
                results.append((xyxy, conf, cls))
    return results

def detect_objects(image_path):
    dicom_image, dicom_meta = read_and_preprocess_dicom(image_path)
    img, img0 = preprocess_image(dicom_image)
    pred = infer(model, img)
    results = postprocess(pred, dicom_image, img)
    return results, dicom_image, dicom_meta

def draw_bounding_boxes(img, results, dicom_meta):
    dets = []
    for (x1, y1, x2, y2), conf, cls in results:
        zc = dicom_meta.loc[dicom_meta.Key == 'Instance Number', 'Value'].iloc[0]
        x1, y1, x2, y2, zc, cls = map(int, [x1, y1, x2, y2, zc, cls])
        xc = x1+(x2-x1)/2
        yc = y1+(y2-y1)/2
        conf = round(conf.detach().item(), 4)

        dets.append([(xc, yc, zc), conf, cls])
        cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
        cv2.putText(img, f'{model.names[int(cls)]} {conf:.2f}', (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (36, 255, 12), 2)
    return img, dets

def show_preds_image(filepath):
    results, img0, dicom_meta = detect_objects(filepath)
    img_with_boxes, results = draw_bounding_boxes(img0, results, dicom_meta)
    print("Detections:", dicom_meta.loc[dicom_meta.Key == 'Series Instance UID', 'Value'].iloc[0], results)
    return cv2.cvtColor(img_with_boxes, cv2.COLOR_BGR2RGB), results, dicom_meta

def read_and_preprocess_dicom(file_path: str):
    """
    Function to read and preprocess DICOM files
    :param file_path: Path to the DICOM file
    :return: Image data (in CV2 format) and metadata (in pandas DataFrame format)
    """
    try:
        # Read the DICOM file
        dicom_data = pydicom.dcmread(file_path)
    except InvalidDicomError:
        raise gr.Error("The uploaded file is not a valid DICOM file.")

    # Get the pixel data
    try:
        pixel_array = dicom_data.pixel_array
    except AttributeError:
        raise gr.Error("The uploaded DICOM file has no pixel data.")

    # Normalize the pixel data to 8-bit and convert to a PIL image
    if pixel_array.dtype != np.uint8:
        pixel_array = ((pixel_array - np.min(pixel_array)) / (np.max(pixel_array) - np.min(pixel_array)) * 255).astype(
            np.uint8)
    image_pil = Image.fromarray(pixel_array)
    
    image = image_pil.convert('RGB')

    image = np.array(image)[:,:,::-1].copy()

    # Collect metadata in dictionary format and convert to DataFrame
    metadata_dict = {elem.name: str(elem.value) for elem in dicom_data.iterall() if elem.name != 'Pixel Data'}
    df_metadata = pl.DataFrame({
        "Key": list(metadata_dict.keys()),
        "Value": list(metadata_dict.values())
    })

    return image, df_metadata.to_pandas()  # Convert to pandas DataFrame for Gradio compatibility


# Define Gradio components
input_component = gr.File(label="Input DICOM Data")
dicom_image = gr.Image(type="numpy", label="Output Image")
dicom_meta = gr.Dataframe(headers=None, label="Metadata")
dets_res = gr.Text(label="Detections")
output_component = [dicom_image, dets_res, dicom_meta]

# Create Gradio interface
interface = gr.Interface(
    fn=show_preds_image,
    inputs=input_component,
    outputs=output_component,
    title="Lung Nodule Detection",
    examples=['samples/110_109.dcm','samples/189_188.dcm'],
    description= "This online deployment proves the effectiveness and efficient function of the machine learning model in identifying lung cancer nodules.",
    live=False,
)

interface.launch()