Skip to content

Latest commit

 

History

History
155 lines (75 loc) · 3.83 KB

RP3D_Demo_Instruction.md

File metadata and controls

155 lines (75 loc) · 3.83 KB
Raw

Instructions for RP3D_Demo

Here we give the instructions to run RP3D_Demo we provided from scratch.

Example Data

Here we provide 3 cases formatted as .npy files in .../RP3D_Demo/DataPath/DemoData/

The data info file is .../RP3D_Demo/DataPath/demoData.json

Case 1: This case contains one CT volume and one MRI volume. For more intuitionistic information, please refer to Radiopaedia Case 1

Case 2: This case contains one X-ray image and two CT volume. For more intuitionistic information, please refer to Radiopaedia Case 2

Case 3: This case contains one X-ray image and two MRI volumes. For more intuitionistic information, please refer to Radiopaedia Case3

Environment Setup

Step1:

pip install -r requirements

Please refer to requirements.txt for detailed information

Step2:

Download the model checkpoint folder RP3D_5569_res_ART_256_32_BCE_T_4_2_0.7_2048_E_F_F_F_N

Then add this folder to the the Path: .../RP3D_Demo/Logout/

Files Modification

Please modify all the relative paths in files to the absolute path in your machine. These files include:

.../RP3D_Demo/Model/modelRadNet.py: Change all the sys path to fit your environment.

.../RP3D_Demo/Model/vit2D.py: Change all the sys path to fit your environment.

.../RP3D_Demo/Model/vitFuse.py: Change all the sys path to fit your environment.

.../RP3D_Demo/eval.py: Change all the sys path to fit your environment.

.../RP3D_Demo/eval.sh: Change all the sys path to fit your environment.

.../RP3D_Demo/verify.py: Change all the sys path to fit your environment.

Running Process

In the beginning , all the required files will form the structure in the following

.../RP3D_Demo

|-- DataOutput

|-- DataPath

​ |-- DemoData

​ |-- data

​ |--key

​ |-- demoData.json

​ |-- disorder_label_dict.json

|-- Datset

​ |-- dataset.py

|-- Logout

​ |-- RP3D_5569_res_ART_256_32_BCE_T_4_2_0.7_2048_E_F_F_F_N

​ |-- log0Eval

|-- Loss

​ |-- AllLosses.py

|-- Model

​ |-- trainsformer_encoder

​ |-- modelRadnet.py

​ |-- position_encoding.py

​ |-- resAttention.py

​ |-- resnet2D.py

​ |-- vit2D.py

​ |-- vitFuse.py

|-- Utils

​ |-- utils.py

|-- eval.py

|-- eval.sh

|-- verify.py

In eval.py, since there is no training process, we can set train_path = eval_path = '.../RP3D_Demo/DataPath/demoData.json'

in eval.sh, keep the parameter FUSE="early".

Then run in terminal:

cd .../RP3D_Demo
bash eval.sh

After running, there will be two files generated, they are:

.../RP3D_Demo/DataOutput/eval_logits.npy: the model's prediction logits.

.../RP3D_Demo/DataOutput/eval_labels.npy: corresponding label.

Then run in terminal:

python .../RP3D_Demo/verify.py

The final output is:

Case 0, gt disorders: ['Abdominal aortic aneurysm', 'Abdominal aortic aneurysm rupture'], predict result: [True, True], predict score: [0.6873902082443237, 0.5604543089866638], corresponding thresholds: [0.3722658157348633, 0.14874456822872162]

Case 1, gt disorders: ['Hip dislocation'], predict result: [True], predict score: [0.5709783434867859], corresponding thresholds: [0.2768756449222564]

Case 2, gt disorders: ['Anterior cruciate ligament tear', 'Segond fracture'], predict result: [True, True], predict score: [0.24337439239025116, 0.27158036828041077], corresponding thresholds: [0.094383917748928, 0.179479107260704]