STELLAR-CLASS-5L-6 (Custom CNN) - Fetal Planes Classification

This model is a custom 5-block convolutional network, trained from scratch (no ImageNet pretraining) on single-channel 128x128 fetal ultrasound images, to classify fetal ultrasound standard planes.

Model Performance (Test Set)

• Accuracy: 0.9365
• Macro F1 Score: 0.9258
• Macro Precision: 0.9080
• Macro Recall: 0.9477

---

Training History

Training & Evaluation Metrics Summary

Overall Training Job Metrics

• Total Epochs Completed: 77.0
• Total Training Steps: 14,861
• Total Training Runtime: 6,722.42 seconds (~1.87 hours)
• Training Throughput: 114.84 samples/sec (2.871 steps/sec)
• Final Overall Training Loss: 0.3823

---

Epoch-by-Epoch Evaluation History

---

Periodic Intermediate Training Log (Every 500 Steps)

---

Key Highlights & Performance Analysis

1. Best Validation Loss:
   • Step 14,089 (Epoch 73.0): Achieved an evaluation loss of 0.1603 with an accuracy of 94.14% and Macro F1 of 0.9349.
2. Best Overall Accuracy & F1 Score:
   • Step 3,860 (Epoch 20.0): Reached a peak accuracy spike of 95.31% and Macro F1 of 0.9461.
   • Late Convergence Plateau: From Epoch 63 onwards, evaluation accuracy consistently held high around 93.0% – 94.1% with minimal loss fluctuations.
3. Training Loss Convergence:
   • Training loss decreased steadily from 1.1721 (Step 500) down to 0.2161 (Step 14,500), showing smooth convergence aligned with the learning rate schedule.

Dataset Information

This model was trained on dataset Marc-HealthAI/fetal-planes-classification-dataset-main.

• Classes: Fetal abdomen, Fetal brain, Fetal femur, Fetal thorax, Maternal cervix, Other
• Dataset Features: ['Image_name', 'Patient_num', 'Plane', 'Brain_plane', 'Operator', 'US_Machine', 'Split', 'image']

Original Dataset README Content

Click to expand original dataset README

Fetal_Planes_DB

Burgos-Artizzu, X.P., Coronado-Gutiérrez, D., Valenzuela-Alcaraz, B. et al. Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes. Sci Rep 10, 10200 (2020). https://doi.org/10.1038/s41598-020-67076-5

Data Description

A large dataset of routinely acquired maternal-fetal screening ultrasound images collected from two different hospitals by several operators and ultrasound machines. All images were manually labeled by an expert maternal fetal clinician (B.V-A.). Images were divided into 6 classes: four of the most widely used fetal anatomical planes (Abdomen, Brain, Femur and Thorax), the mother’s cervix (widely used for prematurity screening) and a general category to include any other less common image plane. Fetal brain images were further categorized into the 3 most common fetal brain planes (Trans-thalamic, Trans-cerebellum, Trans-ventricular) to judge fine grain categorization performance. The final dataset is comprised of over 12,400 images from 1,792 patients.

The dataset details are described in our open-acces paper: Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes

If you find this dataset useful, please cite:

@article{Burgos-ArtizzuFetalPlanesDataset,
  title={Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes},
  author={Burgos-Artizzu, X.P. and Coronado-Gutiérrez, D. and Valenzuela-Alcaraz, B. and Bonet-Carne, E. and Eixarch, E. and Crispi, F. and Gratacós, E.},
  journal={Nature Scientific Reports}, 
  volume={10},
  pages={10200},
  doi="10.1038/s41598-020-67076-5",
  year={2020}
}