|
--- |
|
license: apache-2.0 |
|
base_model: google/vit-base-patch16-224 |
|
tags: |
|
- generated_from_trainer |
|
metrics: |
|
- accuracy |
|
model-index: |
|
- name: vit-base-PICAI |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# vit-base-PICAI |
|
|
|
This model is a fine-tuned version of [google/vit-base-patch16-224](https://huggingface.co/google/vit-base-patch16-224) on the PICAI dataset. |
|
PI-CAI (Prostate Imaging: Cancer AI) is an all-new grand challenge, with over 10,000 carefully-curated prostate MRI exams to validate modern AI algorithms and estimate radiologists’ performance at csPCa detection and diagnosis. Key aspects of the study design have been established in conjunction with an international, multi-disciplinary scientific advisory board (16 experts in prostate AI, radiology and urology) —to unify and standardize present-day guidelines, and to ensure meaningful validation of prostate-AI towards clinical translation (Reinke et al., 2022). |
|
More can be found at the official Grand Channel Website: https://pi-cai.grand-challenge.org |
|
|
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.6043 |
|
- Accuracy: 0.7371 |
|
- Roc Auc: 0.7059 |
|
|
|
## Model description |
|
|
|
More information needed |
|
|
|
## Intended uses & limitations |
|
|
|
This model is just a test of how ViT perform with basic fine tuning over a challengin medical imaging dataset, and also to assess the explanation properties of ViT by looking at attention matrices produced by the model. |
|
|
|
## Training and evaluation data |
|
|
|
More information needed |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 4 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Roc Auc | |
|
|:-------------:|:-----:|:----:|:---------------:|:--------:|:-------:| |
|
| 0.4995 | 0.14 | 50 | 0.5423 | 0.7371 | 0.7072 | |
|
| 0.4729 | 0.29 | 100 | 0.6259 | 0.7314 | 0.7183 | |
|
| 0.5558 | 0.43 | 150 | 0.5564 | 0.7243 | 0.7189 | |
|
| 0.5825 | 0.57 | 200 | 0.5912 | 0.6943 | 0.7177 | |
|
| 0.5091 | 0.71 | 250 | 0.5656 | 0.73 | 0.7140 | |
|
| 0.4575 | 0.86 | 300 | 0.5846 | 0.7386 | 0.6858 | |
|
| 0.5168 | 1.0 | 350 | 0.5363 | 0.7471 | 0.7076 | |
|
| 0.5305 | 1.14 | 400 | 0.5600 | 0.7357 | 0.7042 | |
|
| 0.4275 | 1.29 | 450 | 0.5864 | 0.7357 | 0.6988 | |
|
| 0.5588 | 1.43 | 500 | 0.5477 | 0.75 | 0.7078 | |
|
| 0.4573 | 1.57 | 550 | 0.5321 | 0.7571 | 0.7253 | |
|
| 0.5094 | 1.71 | 600 | 0.5840 | 0.7457 | 0.7054 | |
|
| 0.5311 | 1.86 | 650 | 0.5719 | 0.7229 | 0.7098 | |
|
| 0.4582 | 2.0 | 700 | 0.5439 | 0.7357 | 0.7062 | |
|
| 0.5142 | 2.14 | 750 | 0.6668 | 0.6629 | 0.6899 | |
|
| 0.3833 | 2.29 | 800 | 0.5705 | 0.7286 | 0.6954 | |
|
| 0.4676 | 2.43 | 850 | 0.6152 | 0.6943 | 0.6795 | |
|
| 0.4682 | 2.57 | 900 | 0.5679 | 0.7443 | 0.7077 | |
|
| 0.4112 | 2.71 | 950 | 0.5600 | 0.7329 | 0.7073 | |
|
| 0.5107 | 2.86 | 1000 | 0.5686 | 0.7343 | 0.7017 | |
|
| 0.4078 | 3.0 | 1050 | 0.6165 | 0.7429 | 0.7168 | |
|
| 0.479 | 3.14 | 1100 | 0.5952 | 0.7257 | 0.7004 | |
|
| 0.3704 | 3.29 | 1150 | 0.5937 | 0.7314 | 0.6980 | |
|
| 0.3733 | 3.43 | 1200 | 0.5923 | 0.7214 | 0.7001 | |
|
| 0.3682 | 3.57 | 1250 | 0.6183 | 0.7429 | 0.6963 | |
|
| 0.3283 | 3.71 | 1300 | 0.6130 | 0.73 | 0.7012 | |
|
| 0.3709 | 3.86 | 1350 | 0.6123 | 0.74 | 0.7045 | |
|
| 0.3859 | 4.0 | 1400 | 0.6043 | 0.7371 | 0.7059 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.39.1 |
|
- Pytorch 2.2.1+cu121 |
|
- Datasets 2.18.0 |
|
- Tokenizers 0.15.2 |
|
|
