AnkleNet: ankle-specific model for chronic ankle instability diagnosis

Background

Chronic ankle instability (CAI), characterized by decreased ankle stability and recurrent injuries, occurs in around 40% of ankle sprain patients. The most severe complication of CAI is end-stage traumatic arthritis, which cannot be completely restored through surgical interventions. Therefore, prompt diagnosis and early intervention are essential.

AnkleNet

To facilitate the diagnosis of CAI, we developed a transformer-based model, named AnkleNet. It can detect the injuries of lateral and medial collateral ligaments simultaneously based on MRI, aiding classifying of CAI patients in a detailed way: Normal, LCAI, MCAI, and RCAI.

To train the model, follow the steps below.

Step 1

Preprocess your MRI and make a csv files of your (images label) pairs.

The demo csv files can be found in data/.

• AxialPath: the image path of axial mri

• CoronalPath: the image path of coronal mri

• label1: lateral collateral ligament injury

• label2: medial collateral ligament injury

Step 2

Modify your training configs.

The config templete can be cound in config/.

Step 3

Train the model. Run:

python main_train.py --opt config/anklenet.yaml

---

Acknowlegements

• vit-pytorch (https://github.com/lucidrains/vit-pytorch)

• pytorch-grad-cam (https://github.com/jacobgil/pytorch-grad-cam)