Intracranial_Hemorrhage / Git / Diff of /README.md

Models:

DavidFeaster/

Intracranial_Hemorrhage

Downloads: 1

Diff of /README.md [9f60b7] .. [2bae97]

Switch to unified view


# A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans

[Journal Link](https://doi.org/10.1016/j.nicl.2021.102785)

# RSNA Intracranial Hemorrhage Detection
This is the source code for the first place solution to the [RSNA2019 Intracranial Hemorrhage Detection Challenge](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection).

Solution write up: [Link](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/117210#latest-682640).

## Solutuoin Overview
![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/overview.png?raw=true)

#### Dependencies
- opencv-python==3.4.2
- scikit-image==0.14.0
- scikit-learn==0.19.1
- scipy==1.1.0
- torch==1.1.0
- torchvision==0.2.1

### CODE
- 2DNet
- 3DNet
- SequenceModel

# 2D CNN Classifier

## Pretrained models
- seresnext101_256*256 [\[seresnext101\]](https://drive.google.com/open?id=18Py5eW1E4hSbTT6658IAjQjJGS28grdx)
- densenet169_256*256 [\[densenet169\]](https://drive.google.com/open?id=1vCsX12pMZxBmuGGNVnjFFiZ-5u5vD-h6)
- densenet121_512*512 [\[densenet121\]](https://drive.google.com/open?id=1o0ok-6I2hY1ygSWdZOKmSD84FsEpgDaa)

## Preprocessing
![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/preprocessing.png?raw=true)

Prepare csv file:

download data.zip:  https://drive.google.com/open?id=1buISR_b3HQDU4KeNc_DmvKTYJ1gvj5-3

1. convert dcm to png
```
python3 prepare_data.py -dcm_path stage_1_train_images -png_path train_png
python3 prepare_data.py -dcm_path stage_1_test_images -png_path train_png
python3 prepare_data.py -dcm_path stage_2_test_images -png_path test_png
```

2. train

```
python3 train_model.py -backbone DenseNet121_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet121_change_avg_256
python3 train_model.py -backbone DenseNet169_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet169_change_avg_256
python3 train_model.py -backbone se_resnext101_32x4d -img_size 256 -tbs 80 -vbs 40 -save_path se_resnext101_32x4d_256
```

3. predict
```
python3 predict.py -backbone DenseNet121_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet121_change_avg_256
python3 predict.py -backbone DenseNet169_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet169_change_avg_256
python3 predict.py -backbone se_resnext101_32x4d -img_size 256 -tbs 4 -vbs 4 -spth se_resnext101_32x4d_256
```

After single models training,  the oof files will be saved in ./SingleModelOutput(three folders for three pipelines). 

After training the sequence model, the final submission will be ./FinalSubmission/final_version/submission_tta.csv

# Sequence Models

## Sequence Model 1
![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s1.png)

## Sequence Model 2
![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s2.png)

#### Path Setup
Set data path in ./setting.py

#### download 

download [\[csv.zip\]](https://drive.google.com/open?id=1qYi4k-DuOLJmyZ7uYYrnomU2U7MrYRBV)

download [\[feature samples\]](https://drive.google.com/open?id=1lJgzZoHFu6HI4JBktkGY3qMk--28IUkC)

#### Sequence Model Training
```
CUDA_VISIBLE_DEVICES=0 python main.py
```
The final submissions are in the folder ../FinalSubmission/version2/submission_tta.csv

## Final Submission
### Private Leaderboard:
- 0.04383
## Reference
If you find our work useful in your research or if you use parts of this code please consider citing our [paper](https://doi.org/10.1016/j.nicl.2021.102785):

```@article{wang2021deep,
  title={A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans},
  author={Wang, Xiyue and Shen, Tao and Yang, Sen and Lan, Jun and Xu, Yanming and Wang, Minghui and Zhang, Jing and Han, Xiao},
  journal={NeuroImage: Clinical},
  volume={32},
  pages={102785},
  year={2021},
  publisher={Elsevier}
} 
```


### TODO
- [ ] Pre-trained models
- [ ] 2DCNN + SeqModel end-to-end training 
- [ ] 3DCNN training

	a/README.md		b/README.md
1	# A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans	1	# A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans
2		2
3	[Journal Link](https://doi.org/10.1016/j.nicl.2021.102785)	3	[Journal Link](https://doi.org/10.1016/j.nicl.2021.102785)
4		4
5	# RSNA Intracranial Hemorrhage Detection	5	# RSNA Intracranial Hemorrhage Detection
6	This is the source code for the first place solution to the [RSNA2019 Intracranial Hemorrhage Detection Challenge](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection).	6	This is the source code for the first place solution to the [RSNA2019 Intracranial Hemorrhage Detection Challenge](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection).
7		7
8	Solution write up: [Link](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/117210#latest-682640).	8	Solution write up: [Link](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/117210#latest-682640).
9		9
10	## Solutuoin Overview	10	## Solutuoin Overview
11	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/overview.png)	11	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/overview.png?raw=true)
12		12
13	#### Dependencies	13	#### Dependencies
14	- opencv-python==3.4.2	14	- opencv-python==3.4.2
15	- scikit-image==0.14.0	15	- scikit-image==0.14.0
16	- scikit-learn==0.19.1	16	- scikit-learn==0.19.1
17	- scipy==1.1.0	17	- scipy==1.1.0
18	- torch==1.1.0	18	- torch==1.1.0
19	- torchvision==0.2.1	19	- torchvision==0.2.1
20		20
21	### CODE	21	### CODE
22	- 2DNet	22	- 2DNet
23	- 3DNet	23	- 3DNet
24	- SequenceModel	24	- SequenceModel
25		25
26	# 2D CNN Classifier	26	# 2D CNN Classifier
27		27
28	## Pretrained models	28	## Pretrained models
29	- seresnext101_256*256 [\[seresnext101\]](https://drive.google.com/open?id=18Py5eW1E4hSbTT6658IAjQjJGS28grdx)	29	- seresnext101_256*256 [\[seresnext101\]](https://drive.google.com/open?id=18Py5eW1E4hSbTT6658IAjQjJGS28grdx)
30	- densenet169_256*256 [\[densenet169\]](https://drive.google.com/open?id=1vCsX12pMZxBmuGGNVnjFFiZ-5u5vD-h6)	30	- densenet169_256*256 [\[densenet169\]](https://drive.google.com/open?id=1vCsX12pMZxBmuGGNVnjFFiZ-5u5vD-h6)
31	- densenet121_512*512 [\[densenet121\]](https://drive.google.com/open?id=1o0ok-6I2hY1ygSWdZOKmSD84FsEpgDaa)	31	- densenet121_512*512 [\[densenet121\]](https://drive.google.com/open?id=1o0ok-6I2hY1ygSWdZOKmSD84FsEpgDaa)
32		32
33	## Preprocessing	33	## Preprocessing
34	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/preprocessing.png)	34	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/preprocessing.png?raw=true)
35		35
36	Prepare csv file:	36	Prepare csv file:
37		37
38	download data.zip: https://drive.google.com/open?id=1buISR_b3HQDU4KeNc_DmvKTYJ1gvj5-3	38	download data.zip: https://drive.google.com/open?id=1buISR_b3HQDU4KeNc_DmvKTYJ1gvj5-3
39		39
40	1. convert dcm to png	40	1. convert dcm to png
41	```	41	```
42	python3 prepare_data.py -dcm_path stage_1_train_images -png_path train_png	42	python3 prepare_data.py -dcm_path stage_1_train_images -png_path train_png
43	python3 prepare_data.py -dcm_path stage_1_test_images -png_path train_png	43	python3 prepare_data.py -dcm_path stage_1_test_images -png_path train_png
44	python3 prepare_data.py -dcm_path stage_2_test_images -png_path test_png	44	python3 prepare_data.py -dcm_path stage_2_test_images -png_path test_png
45	```	45	```
46		46
47	2. train	47	2. train
48		48
49	```	49	```
50	python3 train_model.py -backbone DenseNet121_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet121_change_avg_256	50	python3 train_model.py -backbone DenseNet121_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet121_change_avg_256
51	python3 train_model.py -backbone DenseNet169_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet169_change_avg_256	51	python3 train_model.py -backbone DenseNet169_change_avg -img_size 256 -tbs 256 -vbs 128 -save_path DenseNet169_change_avg_256
52	python3 train_model.py -backbone se_resnext101_32x4d -img_size 256 -tbs 80 -vbs 40 -save_path se_resnext101_32x4d_256	52	python3 train_model.py -backbone se_resnext101_32x4d -img_size 256 -tbs 80 -vbs 40 -save_path se_resnext101_32x4d_256
53	```	53	```
54		54
55	3. predict	55	3. predict
56	```	56	```
57	python3 predict.py -backbone DenseNet121_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet121_change_avg_256	57	python3 predict.py -backbone DenseNet121_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet121_change_avg_256
58	python3 predict.py -backbone DenseNet169_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet169_change_avg_256	58	python3 predict.py -backbone DenseNet169_change_avg -img_size 256 -tbs 4 -vbs 4 -spth DenseNet169_change_avg_256
59	python3 predict.py -backbone se_resnext101_32x4d -img_size 256 -tbs 4 -vbs 4 -spth se_resnext101_32x4d_256	59	python3 predict.py -backbone se_resnext101_32x4d -img_size 256 -tbs 4 -vbs 4 -spth se_resnext101_32x4d_256
60	```	60	```
61		61
62	After single models training, the oof files will be saved in ./SingleModelOutput(three folders for three pipelines).	62	After single models training, the oof files will be saved in ./SingleModelOutput(three folders for three pipelines).
63		63
64	After training the sequence model, the final submission will be ./FinalSubmission/final_version/submission_tta.csv	64	After training the sequence model, the final submission will be ./FinalSubmission/final_version/submission_tta.csv
65		65
66	# Sequence Models	66	# Sequence Models
67		67
68	## Sequence Model 1	68	## Sequence Model 1
69	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s1.png)	69	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s1.png)
70		70
71	## Sequence Model 2	71	## Sequence Model 2
72	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s2.png)	72	![image](https://github.com/SeuTao/RSNA2019_1st_place_solution/blob/master/docs/s2.png)
73		73
74	#### Path Setup	74	#### Path Setup
75	Set data path in ./setting.py	75	Set data path in ./setting.py
76		76
77	#### download	77	#### download
78		78
79	download [\[csv.zip\]](https://drive.google.com/open?id=1qYi4k-DuOLJmyZ7uYYrnomU2U7MrYRBV)	79	download [\[csv.zip\]](https://drive.google.com/open?id=1qYi4k-DuOLJmyZ7uYYrnomU2U7MrYRBV)
80		80
81	download [\[feature samples\]](https://drive.google.com/open?id=1lJgzZoHFu6HI4JBktkGY3qMk--28IUkC)	81	download [\[feature samples\]](https://drive.google.com/open?id=1lJgzZoHFu6HI4JBktkGY3qMk--28IUkC)
82		82
83	#### Sequence Model Training	83	#### Sequence Model Training
84	```	84	```
85	CUDA_VISIBLE_DEVICES=0 python main.py	85	CUDA_VISIBLE_DEVICES=0 python main.py
86	```	86	```
87	The final submissions are in the folder ../FinalSubmission/version2/submission_tta.csv	87	The final submissions are in the folder ../FinalSubmission/version2/submission_tta.csv
88		88
89	## Final Submission	89	## Final Submission
90	### Private Leaderboard:	90	### Private Leaderboard:
91	- 0.04383	91	- 0.04383
92	## Reference	92	## Reference
93	If you find our work useful in your research or if you use parts of this code please consider citing our [paper](https://doi.org/10.1016/j.nicl.2021.102785):	93	If you find our work useful in your research or if you use parts of this code please consider citing our [paper](https://doi.org/10.1016/j.nicl.2021.102785):
94		94
95	```@article{wang2021deep,	95	```@article{wang2021deep,
96	title={A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans},	96	title={A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans},
97	author={Wang, Xiyue and Shen, Tao and Yang, Sen and Lan, Jun and Xu, Yanming and Wang, Minghui and Zhang, Jing and Han, Xiao},	97	author={Wang, Xiyue and Shen, Tao and Yang, Sen and Lan, Jun and Xu, Yanming and Wang, Minghui and Zhang, Jing and Han, Xiao},
98	journal={NeuroImage: Clinical},	98	journal={NeuroImage: Clinical},
99	volume={32},	99	volume={32},
100	pages={102785},	100	pages={102785},
101	year={2021},	101	year={2021},
102	publisher={Elsevier}	102	publisher={Elsevier}
103	}	103	}
104	```	104	```
105		105
106		106
107	### TODO	107	### TODO
108	- [ ] Pre-trained models	108	- [ ] Pre-trained models
109	- [ ] 2DCNN + SeqModel end-to-end training	109	- [ ] 2DCNN + SeqModel end-to-end training
110	- [ ] 3DCNN training	110	- [ ] 3DCNN training