Datasets
Models
Downloads: 1
[dc3c86]:
/
MRNet_fastai_example.ipynb
800 lines (799 with data), 47.3 kB
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import pandas as pd\n",
"import matplotlib.pyplot as plt\n",
"from fastai.vision import *\n",
"import torch\n",
"\n",
"from mrnet_orig import *\n",
"\n",
"%matplotlib inline"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[01;34m..\u001b[00m\r\n",
"├── \u001b[01;34mdata\u001b[00m\r\n",
"│ ├── \u001b[01;34maxial\u001b[00m\r\n",
"│ │ ├── \u001b[01;34mtrain\u001b[00m\r\n",
"│ │ └── \u001b[01;34mvalid\u001b[00m\r\n",
"│ ├── \u001b[01;34mcoronal\u001b[00m\r\n",
"│ │ ├── \u001b[01;34mtrain\u001b[00m\r\n",
"│ │ └── \u001b[01;34mvalid\u001b[00m\r\n",
"│ └── \u001b[01;34msagittal\u001b[00m\r\n",
"│ ├── \u001b[01;34mmodels\u001b[00m\r\n",
"│ ├── \u001b[01;34mtrain\u001b[00m\r\n",
"│ └── \u001b[01;34mvalid\u001b[00m\r\n",
"└── \u001b[01;34mmrnet-fastai\u001b[00m\r\n",
" ├── \u001b[01;34mexp\u001b[00m\r\n",
" └── \u001b[01;34m__pycache__\u001b[00m\r\n",
"\r\n",
"14 directories\r\n"
]
}
],
"source": [
"! tree -d .."
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"df_abnl.pkl loss_weights.pt\t\t mrnet_orig.py\tslice_stats.json\r\n",
"exp\t MRNet_EDA.ipynb\t\t __pycache__\ttrain_cases.pkl\r\n",
"LICENSE MRNet_fastai_example.ipynb README.md\ttrain_pix_distr.pkl\r\n"
]
}
],
"source": [
"! ls"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"axial\t train-abnormal.csv valid-abnormal.csv\r\n",
"coronal train-acl.csv valid-acl.csv\r\n",
"sagittal train-meniscus.csv valid-meniscus.csv\r\n"
]
}
],
"source": [
"! ls ../data"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"data_path = Path('../data')\n",
"sag_path = data_path/'sagittal'\n",
"cor_path = data_path/'coronal'\n",
"ax_path = data_path/'axial'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Substantial class imbalance for the normal/abnormal task\n",
"\n",
"Given this, we'll derive weights for a weighted binary cross entropy loss function."
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(1130, 2)\n"
]
},
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Case</th>\n",
" <th>Abnormal</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0000</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>0001</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>0002</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>0003</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>0004</td>\n",
" <td>1</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Case Abnormal\n",
"0 0000 1\n",
"1 0001 1\n",
"2 0002 1\n",
"3 0003 1\n",
"4 0004 1"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"train_abnl = pd.read_csv(data_path/'train-abnormal.csv', header=None,\n",
" names=['Case', 'Abnormal'], \n",
" dtype={'Case': str, 'Abnormal': np.int64})\n",
"print(train_abnl.shape)\n",
"train_abnl.head()"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0.8079646017699115\n",
"tensor([0.8080, 0.1920])\n"
]
}
],
"source": [
"w = train_abnl.Abnormal.sum() / train_abnl.shape[0]\n",
"print(w)\n",
"weights = Tensor([w, 1-w])\n",
"print(weights)\n",
"torch.save(weights, 'loss_weights.pt')"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"weights = torch.load('loss_weights.pt')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Load previously created files\n",
"\n",
"- `df_abnl` -> master `df` for use with Data Block API, also contains # of slices per series\n",
"- `slice_stats` -> `dict` stored as `json` with mean and max # of slices per series"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Case</th>\n",
" <th>Abnormal</th>\n",
" <th>is_valid</th>\n",
" <th>coronal_slices</th>\n",
" <th>sagittal_slices</th>\n",
" <th>axial_slices</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>train/0000</td>\n",
" <td>1</td>\n",
" <td>0</td>\n",
" <td>25</td>\n",
" <td>27</td>\n",
" <td>25</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>train/0001</td>\n",
" <td>1</td>\n",
" <td>0</td>\n",
" <td>22</td>\n",
" <td>23</td>\n",
" <td>28</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>train/0002</td>\n",
" <td>1</td>\n",
" <td>0</td>\n",
" <td>24</td>\n",
" <td>24</td>\n",
" <td>24</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>train/0003</td>\n",
" <td>1</td>\n",
" <td>0</td>\n",
" <td>22</td>\n",
" <td>21</td>\n",
" <td>25</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>train/0004</td>\n",
" <td>1</td>\n",
" <td>0</td>\n",
" <td>30</td>\n",
" <td>30</td>\n",
" <td>31</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Case Abnormal is_valid coronal_slices sagittal_slices \\\n",
"0 train/0000 1 0 25 27 \n",
"1 train/0001 1 0 22 23 \n",
"2 train/0002 1 0 24 24 \n",
"3 train/0003 1 0 22 21 \n",
"4 train/0004 1 0 30 30 \n",
"\n",
" axial_slices \n",
"0 25 \n",
"1 28 \n",
"2 24 \n",
"3 25 \n",
"4 31 "
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df_abnl = pd.read_pickle('df_abnl.pkl')\n",
"df_abnl.head()"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'coronal': {'mean': 29.6416, 'max': 57},\n",
" 'sagittal': {'mean': 30.3776, 'max': 51},\n",
" 'axial': {'mean': 34.2032, 'max': 61}}"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"with open('slice_stats.json', 'r') as file:\n",
" stats = json.load(file)\n",
" \n",
"stats"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"51\n"
]
}
],
"source": [
"max_slc = stats['sagittal']['max']\n",
"print(max_slc)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## MRNet implementation\n",
"\n",
"Modified from the original [paper](https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002699) to (sort of) work with `fastai`"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [],
"source": [
"il = MR3DImageList.from_df(df_abnl, sag_path, suffix='.npy')"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'../data/sagittal/train/0000.npy'"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"il.items[0]"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"MR3DImageList (1250 items)\n",
"Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256)\n",
"Path: ../data/sagittal"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"il"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"ItemLists;\n",
"\n",
"Train: MR3DImageList (1130 items)\n",
"Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256)\n",
"Path: ../data/sagittal;\n",
"\n",
"Valid: MR3DImageList (120 items)\n",
"Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256)\n",
"Path: ../data/sagittal;\n",
"\n",
"Test: None"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"sd = il.split_from_df(col=2)\n",
"sd"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"LabelLists;\n",
"\n",
"Train: LabelList (1130 items)\n",
"x: MR3DImageList\n",
"Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256)\n",
"y: CategoryList\n",
"1,1,1,1,1\n",
"Path: ../data/sagittal;\n",
"\n",
"Valid: LabelList (120 items)\n",
"x: MR3DImageList\n",
"Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256),Image (51, 3, 256, 256)\n",
"y: CategoryList\n",
"0,0,0,0,0\n",
"Path: ../data/sagittal;\n",
"\n",
"Test: None"
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"ll = sd.label_from_df(cols=1)\n",
"ll"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [],
"source": [
"# tfms = get_transforms()"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [],
"source": [
"bs = 1\n",
"data = ll.databunch(bs=bs)"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [],
"source": [
"learn = mrnet_learner(data, MRNet(), opt_func=optim.Adam, loss_func=WtBCELoss(weights),\n",
" callbacks=MRNetCallback(), metrics=accuracy)"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"======================================================================\n",
"Layer (type) Output Shape Param # Trainable \n",
"======================================================================\n",
"Conv2d [64, 63, 63] 23,296 False \n",
"______________________________________________________________________\n",
"ReLU [64, 63, 63] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [64, 31, 31] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [192, 31, 31] 307,392 False \n",
"______________________________________________________________________\n",
"ReLU [192, 31, 31] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [192, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [384, 15, 15] 663,936 False \n",
"______________________________________________________________________\n",
"ReLU [384, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [256, 15, 15] 884,992 False \n",
"______________________________________________________________________\n",
"ReLU [256, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [256, 15, 15] 590,080 False \n",
"______________________________________________________________________\n",
"ReLU [256, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [256, 7, 7] 0 False \n",
"______________________________________________________________________\n",
"AdaptiveAvgPool2d [256, 1, 1] 0 False \n",
"______________________________________________________________________\n",
"Linear [1] 257 True \n",
"______________________________________________________________________\n",
"\n",
"Total params: 2,469,953\n",
"Total trainable params: 257\n",
"Total non-trainable params: 2,469,696"
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"learn.summary()"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
]
}
],
"source": [
"learn.lr_find()"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"learn.recorder.plot()"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: left;\">\n",
" <th>epoch</th>\n",
" <th>train_loss</th>\n",
" <th>valid_loss</th>\n",
" <th>accuracy</th>\n",
" <th>time</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <td>0</td>\n",
" <td>0.443489</td>\n",
" <td>0.453011</td>\n",
" <td>0.208333</td>\n",
" <td>08:15</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>"
],
"text/plain": [
"<IPython.core.display.HTML object>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"learn.fit_one_cycle(1, 3e-4)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Accuracy is terrible, but what do you expect out of a single linear layer...?"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [],
"source": [
"learn.unfreeze()"
]
},
{
"cell_type": "code",
"execution_count": 25,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"======================================================================\n",
"Layer (type) Output Shape Param # Trainable \n",
"======================================================================\n",
"Conv2d [64, 63, 63] 23,296 True \n",
"______________________________________________________________________\n",
"ReLU [64, 63, 63] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [64, 31, 31] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [192, 31, 31] 307,392 True \n",
"______________________________________________________________________\n",
"ReLU [192, 31, 31] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [192, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [384, 15, 15] 663,936 True \n",
"______________________________________________________________________\n",
"ReLU [384, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [256, 15, 15] 884,992 True \n",
"______________________________________________________________________\n",
"ReLU [256, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"Conv2d [256, 15, 15] 590,080 True \n",
"______________________________________________________________________\n",
"ReLU [256, 15, 15] 0 False \n",
"______________________________________________________________________\n",
"MaxPool2d [256, 7, 7] 0 False \n",
"______________________________________________________________________\n",
"AdaptiveAvgPool2d [256, 1, 1] 0 False \n",
"______________________________________________________________________\n",
"Linear [1] 257 True \n",
"______________________________________________________________________\n",
"\n",
"Total params: 2,469,953\n",
"Total trainable params: 2,469,953\n",
"Total non-trainable params: 0"
]
},
"execution_count": 25,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"learn.summary()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.2"
}
},
"nbformat": 4,
"nbformat_minor": 2
}