McSpooder

    def __getitem__(self, idx):
        
        if torch.is_tensor(idx):
            idx = idx.tolist()
            
        repeat = True                 
    
        while(repeat):
            try:
                path     = self.directories[idx]

McSpooder

class MRIDataset(Dataset):
    
    def __init__(self, root_dir, labels, transform=None):

        self.root_dir = root_dir        
        self.transform = transform     
        self.directories = []
        self.len = 0
        self.labels = labels
        

McSpooder

sMCI_df = clin[clin["label"] == "sMCI"]
pMCI_df = clin[clin["label"] == "pMCI"]

kde_kws = {"color": "orange", "lw": 2, "label": "sMCI", "shade":True}
sns.distplot(sMCI_df["AGE"], kde_kws=kde_kws, hist=False)

kde_kws = {"color": "red", "lw": 2, "label": "pMCI", "shade":True}
sns.distplot(pMCI_df["AGE"], color="magenta", kde_kws=kde_kws, hist=False)

McSpooder

import seaborn as sns
import pandas as pd
import matplotlib as plt

clin = pd.read_csv("./bas_clin_with_categorical.csv")

clin.head()

McSpooder

        out_a     = self.stack3_a(out_a)
        out_b     = self.stack3_b(out_b)
        out       = torch.cat((out_a, out_b), 1)

McSpooder

        identity  = out
        out       = self.layer(out)
        out       = out + identity

McSpooder

class ConvBlock(nn.Module):

    def __init__(self, c_in, c_out, ks, k_stride=1):
        
        super().__init__()

        self.conv1 = nn.Conv3d(c_in, c_out, ks, stride=k_stride, padding=(1,1,1))
        self.bn = nn.BatchNorm3d(c_out)
        self.elu = nn.ELU()
        self.pool = nn.MaxPool3d(kernel_size=(3,3,3), stride=2)

McSpooder

import matplotlib.pyplot as plt
import nibabel as nib

def show_image(image):
    plt.imshow(image)
   
fig = plt.figure()

mri = nib.load("path_to_mri_scan\\mri_volume.nii").get_fdata()
mri.shape

McSpooder

def X_Y_df_split(in_df):
    X_df = in_df.drop(columns=['fun', 'valuable', 'exciting', 'awesome', 'cool'], axis=1)
    Y_df = in_df.drop(columns=['id', 'comment_text'], axis=1)
    return (X_df, Y_df)