test.py

from utils.losses import *
from utils.preprocess import *
from keras.models import load_model

def predict(T1_path,FLAIR_path,IR_path,label):
    model = load_model("weights/label"+str(label)+"/best.h5",custom_objects={'dice_coefficient': dice_coefficient, 'dice_loss':  dice_loss})
    if label in [1,3,5]:
        T1 = get_data_with_skull_scraping(T1_path)
        if label == 5:
            X = np.array((T1>=10) & (T1<110)).astype(np.uint8)[None,None,...]
        elif label == 3:
            X = np.array(T1>=150).astype(np.uint8)[None,None,...]
        else:
            X = np.array((T1>=80) & (T1<160)).astype(np.uint8)[None,None,...]
        y_pred = model.predict(X)
    else:
        T1 = histeq(to_uint8(get_data_with_skull_scraping(T1_path)))[None,None,...]
        IR = IR_to_uint8(get_data(IR_path))[None,None,...]
        FLAIR = to_uint8(get_data(FLAIR_path))[None,None,...]
        y_pred = model.predict([T1,FLAIR,IR])
    return y_pred.squeeze()