---
license: apache-2.0
---
# Thyroid Ultrasonograph Image Classifier

**Author:** Afif Ali Saadman
**Type:** Deep Learning (Modified AlexNet variant)
**Framework:** PyTorch
**Date:** October 2025

## Overview

**This model was developed as part of an **independent research project** focused on classifying thyroid ultrasound images into multiple diagnostic categories using deep learning.**


The model can identify:

- **FTC** – Follicular Thyroid Carcinoma
- **PTC** – Papillary Thyroid Carcinoma
- **MTC** – Medullary Thyroid Carcinoma
- **Benign** – Non-cancerous thyroid tissue


## Architecture

NOTE: This model was trained on a T4 GPU in google colaboratory.

* Model : This model was trained on a AlexNet like architecture with gradient checkpointing.
* Loss Function: Cross Entropy Loss
* Learning Rate: 0.0001
* Iters(Epochs): 20

## Dataset

This dataset  was extracted from `FangDai/Thyroid_Ultrasound_Images `and `agent593/Thyroid-Ultrasound-Image-Classification-ViTModel/tree/main/dataset%20thyroid/` which were cleaned manually.

1. **FTC (Follicular Thyroid Carcinoma) – 100 images**
2. **PTC (Papillary Thyroid Carcinoma) – 99 images**
3. **MTC (Medullary Thyroid Carcinoma) – 99 images**
4. **Benign (Normal Thyroid) - 90 images**

## Confusion Matrix

![Confusion Matrix](matrix.png)


## Classification Report



| Class            | Precision | Recall | F1-Score | Support |
| ---------------- | --------- | ------ | -------- | ------- |
| FTC              | 0.93      | 0.93   | 0.93     | 15      |
| PTC              | 0.88      | 0.70   | 0.78     | 10      |
| MTC              | 0.80      | 0.80   | 0.80     | 10      |
| Benign           | 0.88      | 1.00   | 0.94     | 15      |
| **Accuracy**     | -         | -      | 0.88     | 50      |
| **Macro Avg**    | 0.87      | 0.86   | 0.86     | 50      |
| **Weighted Avg** | 0.88      | 0.88   | 0.88     | 50      |



## Final Report

Benign: perfect classification (15/15)

FTC: only one misclassified

PTC: 2 misclassified (one as FTC, one as Benign)

MTC: also strong, only a few mislabels


## More information

For more information, kindly see this notebook:[USGResearch.ipynb · Afifsudoers/Thyroid-Canciroma-Image-Classifier-Model at main](https://huggingface.co/Afifsudoers/Thyroid-Canciroma-Image-Classifier-Model/blob/main/USGResearch.ipynb)


## Where you can find this model?

**HuggingFace**:[Afifsudoers/Thyroid-Canciroma-Image-Classifier-Model · Hugging Face](https://huggingface.co/Afifsudoers/Thyroid-Canciroma-Image-Classifier-Model)

Kaggle: [Afif Ali Saadman | Thyroid\_Canciroma\_Classifier | Kaggle](https://www.kaggle.com/models/afifalisaadman/thyroid-canciroma-classifier/)


## Citation

```
@misc{saadman2025thyroid,
  author = {Afif Ali Saadman},
  title = {Thyroid Ultrasonograph Image Classifier},
  year = {2025},
  month = {October},
  note = {Deep Learning (Modified AlexNet variant), PyTorch. Available at \url{https://huggingface.co/Afifsudoers/Thyroid-Canciroma-Image-Classifier-Model} and \url{https://www.kaggle.com/models/afifalisaadman/thyroid-canciroma-classifier/}}
}

```