---
license: mit
metrics:
- accuracy: 96.25%
base_model:
- timm/tf_efficientnetv2_b0.in1k
---

## Model Details


### Model Description

- **Developed by:** Arrun Sivasubramanian et. al
- **License:** MIT

## How to Get Started with the Model

Use the code below to get started with the model.
```python


import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import *
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.models import Model
from keras_self_attention import SeqSelfAttention
from sklearn.metrics import *
import seaborn as sns
import matplotlib.pyplot as plt
from tensorflow.keras.applications.efficientnet_v2 import EfficientNetV2B0


input_shape = (sz, sz, 3) # Adjust the shape according to your data
input_layer = Input(shape=input_shape)
base_model = EfficientNetV2B0(include_top=False, weights='imagenet', input_tensor=input_layer).output
x = GlobalAveragePooling2D()(base_model)
x = Dense(64,activation = 'relu')(x)
x = Dense(32,activation = 'relu')(x)
x = Dropout(0.1)(x)
num_classes = 6 # Six, including 'Other' label
output_layer = Dense(num_classes, activation='softmax')(x)

model = Model(inputs=input_layer, outputs=output_layer)
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.load_weights("./EfficientNetV2B0_BestWeights.h5")

data_dir = "./FetalClassification/Split_Images" # https://zenodo.org/records/3904280
train_data_dir = f'{data_dir}/train'
test_data_dir = f'{data_dir}/val'

sz = 256
bs = 16

train_datagen = ImageDataGenerator(
    rotation_range=30,  
    zoom_range=0.2,
    horizontal_flip=True,  
    vertical_flip=True,  
)

test_datagen = ImageDataGenerator()

# Create train and test datasets using image_dataset_from_directory
train_generator  = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(sz,sz),
    batch_size=bs,
    class_mode = 'categorical', 
    seed=42,  # Strictly setting seed for reproducibility
)

test_generator  = test_datagen.flow_from_directory(
    test_data_dir,
    target_size=(sz,sz),
    batch_size=bs,
    class_mode = 'categorical', 
    seed=42,  # Strictly setting seed for reproducibility
    shuffle=False,
)

predictions = model.predict(test_generator)
predicted_labels = np.argmax(predictions,axis = 1)
true_labels = test_generator.classes

classification_report_result = classification_report(true_labels, predicted_labels,digits = 4)
confusion_matrix_result = confusion_matrix(true_labels, predicted_labels)

```

#### Speeds, Sizes, Times

![image/png](https://cdn-uploads.huggingface.co/production/uploads/654891a8b8ac1a89ff2d7d3f/FSa0H9tMKnpBafWNGZWao.png)


#### Metrics

![image/png](https://cdn-uploads.huggingface.co/production/uploads/654891a8b8ac1a89ff2d7d3f/m5hwSZsmywRFMx0vs_Rdh.png)


## Citation
```bibtex
@article{2410.17396,
Author = {Arrun Sivasubramanian and Divya Sasidharan and Sowmya V and Vinayakumar Ravi},
Title = {Efficient Feature Extraction Using Light-Weight CNN Attention-Based Deep Learning Architectures for Ultrasound Fetal Plane Classification},
Year = {2024},
Eprint = {arXiv:2410.17396},
Doi = {10.1007/s13246-025-01566-6},
}
```