# ==========  DATA NOTES  ========== #
#
# Number of "Sample Images" patients = 20
# Number of "Training" patients  = 1397
# Each patient is associated with between, approximately, 100 and 400 DICOM images
# The cancer rate within the training set is 25.91%
# 



import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import os, sys
import glob
import datetime
import dicom
import scipy.ndimage
from skimage import measure, morphology
from sklearn.metrics import log_loss
from mpl_toolkits.mplot3d.art3d import Poly3DCollection


os.chdir('C://Users//Jay//Desktop//DSB2017')


# ==========  CREATE FUNCTIONS USED IN CODE  ========== #
def dicom_to_image(filename):
    dcm = dicom.read_file(filename)
    img = dcm.pixel_array
    img[img == -2000] = 0
    return img

def get_slice_location(dcm):
    return float(dcm[0x0020, 0x1041].value)

# Returns a list of images for that patient_id, in ascending order of Slice Location
def load_patient(patient_id):
    files = glob.glob('Original_Data//sample_images//{}//*.dcm'.format(patient_id))
    imgs = {}
    for f in files:
        dcm = dicom.read_file(f)
        img = dcm.pixel_array
        img[img == -2000] = 0
        sl = get_slice_location(dcm)
        imgs[sl] = img
    sorted_imgs = [x[1] for x in sorted(imgs.items(), key=lambda x: x[0])]
    return sorted_imgs

# This function takes in a single frame from the DICOM and returns a single frame in RGB format.
def normalise(img):
    normed = (img / 14).astype(np.uint8) # Magic number, scaling to create int between 0 and 255
    img2 = np.zeros([*img.shape, 3], dtype=np.uint8)
    for i in range(3):
        img2[:, :, i] = normed
    return img2



# ==========  INVESTIGATE SAMPLE IMAGES  ========== #
for d in os.listdir("Original_Data//sample_images"):
    print("Patient '{}' has {} scans".format(d, len(os.listdir("Original_Data//sample_images//" + d))))
print('----')
print('Total number of patients: {}\nTotal number of DCM files: {}'.format(len(os.listdir("Original_Data//sample_images")), 
        len(glob.glob("Original_Data//sample_images//*//*.dcm"))))
patient_sizes = [len(os.listdir('Original_Data//sample_images//' + d)) for d in os.listdir('Original_Data//sample_images')]
plt.hist(patient_sizes)
plt.ylabel('Number of patients')
plt.xlabel('DICOM files')
plt.title('Histogram of DICOM count per patient')
sizes = [os.path.getsize(dcm)/1000000 for dcm in glob.glob("Original_Data//sample_images//*//*.dcm")]
print('DCM file sizes: min {:.3}MB, max {:.3}MB, avg {:.3}MB, std {:.3}MB'.format(np.min(sizes),
        np.max(sizes), np.mean(sizes), np.std(sizes)))



# ==========  INVESTIGATE TRAINING SET  ========== #													  
df_train = pd.read_csv('Original_Data//stage1_labels//stage1_labels.csv')
df_train.head()
print('Number of training patients: {}'.format(len(df_train)))
print('Cancer rate: {:.4}%'.format(df_train.cancer.mean()*100))



# ==========  CREATE NAIVE SUBMISSION  ========== #
logloss = log_loss(df_train.cancer, np.zeros_like(df_train.cancer) + df_train.cancer.mean())
print('Training logloss is {}'.format(logloss))
sample = pd.read_csv('Original_Data//stage1_sample_submission//stage1_sample_submission.csv')
sample['cancer'] = df_train.cancer.mean()
sample.to_csv('Submissions//naive_submission_'+datetime.datetime.now().strftime("%Y%m%d%H%M")+'.csv', index=False)



# ==========  INVESTIGATE IMAGE PROPERTIES  ========== #
patientID='0a38e7597ca26f9374f8ea2770ba870d'
imageID='4ec5ef19b52ec06a819181e404d37038'

dcmFile = 'Original_Data//sample_images//'+patientID+'//'+imageID+'.dcm'
print('Filename: {}'.format(dcmFile))
dcmFile = dicom.read_file(dcmFile)
dcmFile

img = dcmFile.pixel_array
img[img == -2000] = 0
plt.axis('off')
plt.imshow(img)
plt.show()
plt.axis('off')
plt.imshow(-img) # Invert colors
plt.show()

files = glob.glob('Original_Data/sample_images/*/*.dcm')
f, plots = plt.subplots(4, 5, sharex='col', sharey='row', figsize=(10, 8))
for i in range(20):
    plots[i // 5, i % 5].axis('off')
    plots[i // 5, i % 5].imshow(dicom_to_image(np.random.choice(files)), cmap=plt.cm.bone)
pat = load_patient(patientID)
f, plots = plt.subplots(11, 10, sharex='all', sharey='all', figsize=(10, 11))
plt.title('Sorted Slices of Patient '+patientID)
for i in range(110):
    plots[i // 10, i % 10].axis('off')
    plots[i // 10, i % 10].imshow(pat[i], cmap=plt.cm.bone)
npat = [normalise(p) for p in pat]