https://github.com/sebukpor/multi-cancer-classification
The Multi-Cancer Classification Tool is a deep learning-powered web application designed to classify different classes of cancers based on medical images, providing accurate and fast results to aid medical specialists in diagnosis.
https://github.com/sebukpor/multi-cancer-classification
neural neural-networks tensorflow tensorflowjs transfer-learning
Last synced: 4 months ago
JSON representation
The Multi-Cancer Classification Tool is a deep learning-powered web application designed to classify different classes of cancers based on medical images, providing accurate and fast results to aid medical specialists in diagnosis.
- Host: GitHub
- URL: https://github.com/sebukpor/multi-cancer-classification
- Owner: Sebukpor
- License: mit
- Created: 2024-09-21T08:59:35.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2025-12-04T06:27:28.000Z (7 months ago)
- Last Synced: 2025-12-07T13:20:05.189Z (7 months ago)
- Topics: neural, neural-networks, tensorflow, tensorflowjs, transfer-learning
- Language: HTML
- Homepage: https://sebukpor.github.io/multi-cancer-classification/
- Size: 88.6 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 🧬 Browser-Based Multi-Cancer Classification Framework Using Depthwise Separable Convolutions for Precision Diagnostics
---
**Authors:** Divine Sebukpor, Ikenna Odezuligbo*, Maimuna Nagey, Michael Chukwuka, Oluwamayowa Akinsuyi (Corresponding Author)
**Access Paper :** [here](https://www.mdpi.com/2075-4418/15/23/3066)
* **Test on Web:** [Multi-Cancer Classification – DAS medhub](https://sebukpor.github.io/multi-cancer-classification/)
**Live Demo (Hugging Face Space):** [https://huggingface.co/spaces/Sebukpor/multi-cancer-gradcam](https://huggingface.co/spaces/Sebukpor/multi-cancer-gradcam)
---
## Overview
This repository documents our work: **“Browser-Based Multi-Cancer Classification Framework Using Depthwise Separable Convolutions for Precision Diagnostics.”**
We present a unified, browser-accessible AI system capable of detecting multiple cancer types from medical images, offering fast, accurate, and accessible diagnostic assistance via a web interface.
---
## Multi-Cancer Classification Tool
The Multi-Cancer Classification Tool is a machine-learning-powered web application developed by **DAS medhub**.
Users can upload a medical image and receive a predicted cancer type. The inference runs entirely in the browser using TensorFlow.js—no specialised hardware or server-side inference needed.
### Features
* **Multi-Cancer Classification:** Supports classification for 26 cancer types (see list below).
* **Web-based Inference:** Model runs client-side in the browser.
* **Fast & Efficient:** Delivers predictions within seconds.
* **Accessible:** Tailored for clinicians, researchers, and healthcare professionals.
---
## Model Architecture
* **Base architecture:** Xception (pretrained backbone, fine-tuned).
* **Key components:**
* Depthwise Separable Convolutions for lightweight computation
* Batch Normalization for stability
* Dropout for regularization
* Dense layers for classification
* **Input size:** 224 × 224 pixels
* **Output:** Softmax over 26 cancer classes.
---
## Data Pre-processing & Training
* **Image resizing:** All images scaled to 224 × 224.
* **Normalization:** Pixel values scaled to [0, 1].
* **Augmentation:** Rotation, zooming, and horizontal flips.
## 🛠️ Training Protocol
| Parameter | Value |
|------------------------|-------------------------------|
| Base Model | Xception (ImageNet weights) |
| Input Size | 224 × 224 × 3 |
| Batch Size | 32 |
| Epochs | 21 |
| Optimizer | Adam (lr = 1e⁻⁴) |
| Loss | Categorical Cross-Entropy |
| Augmentation | Rotation (45°), zoom (0.2), horizontal/vertical flip, shift |
| Regularization | Dropout (0.1–0.4), L2 weight decay |
| Fine-tuned Layers | Last 50 layers of Xception |
| Callbacks | EarlyStopping (patience=10), ReduceLROnPlateau (factor=0.2, patience=5) |
> 💡 The model is **class-balanced**: 4,000 training images per class, 500 validation/test per class.
---
## Performance Metrics
| Metric | Value |
| -------------------------- | ------- |
| **Top-1 Accuracy** | 99.85% |
| **Top-5 Accuracy** | 100.00% |
| **Precision (macro avg.)** | 1.00 |
| **Recall (macro avg.)** | 1.00 |
| **F1-score (macro avg.)** | 1.00 |
---
## Confusion Matrix & Loss Curves
---
## 🧫 Classes of Cancer and Imaging Modalities
Each class corresponds to a **specific imaging modality** from a **documented dataset**, ensuring that uploaded images match the model’s expected input type.
---
| Cancer Type | Dataset Source (Kaggle/Figshare) |
|---------------------|--------------------------------------------------------------------------------------------------|
| Brain Tumor | [Figshare – Cheng (2017)](https://figshare.com/articles/dataset/brain_tumor_dataset/1512427) |
| Acute Lymphoblastic Leukemia | [Kaggle – ALL-IDB](https://www.kaggle.com/datasets/mehradaria/leukemia) |
| Breast Cancer | [BreakHis – Spanhol et al.](https://www.kaggle.com/datasets/anaselmasry/breast-cancer-dataset) |
| Cervical Cancer | [SIPaKMeD – Plissiti et al.](https://www.kaggle.com/datasets/prahladmehandiratta/cervical-cancer-largest-dataset-sipakmed) |
| Kidney CT | [CT-Kidney Dataset](https://www.kaggle.com/datasets/nazmul0087/ct-kidney-dataset-normal-cyst-tumor-and-stone) |
| Lung & Colon | [LC25000 – Borkowski et al.](https://www.kaggle.com/datasets/biplobdey/lung-and-colon-cancer) |
| Lymphoma | [Malignant Lymphoma](https://www.kaggle.com/datasets/andrewmvd/malignant-lymphoma-classification)|
| Oral Cancer | [Oral Histopathology](https://www.kaggle.com/datasets/ashenafifasilkebede/dataset) |
---
### 1. Acute Lymphoblastic Leukemia ↪ [Reference](https://www.kaggle.com/datasets/mehradaria/leukemia)
**Imaging Modality:** Microscopy (blood smear / bone marrow)
* **all_benign:** Benign
* **all_early:** Early
* **all_pre:** Pre
* **all_pro:** Pro
---
### 2. Brain Cancer ↪ [Reference](https://figshare.com/articles/dataset/brain_tumor_dataset/1512427)
**Imaging Modality:** MRI – T1-weighted contrast-enhanced (CE-MRI)
* **brain_glioma:** Glioma
* **brain_menin:** Meningioma
* **brain_tumor:** Pituitary Tumor
---
### 3. Breast Cancer ↪ [Reference](https://www.kaggle.com/datasets/anaselmasry/breast-cancer-dataset)
**Imaging Modality:** Histopathology / Microscopy (digital pathology – breast)
* **breast_benign:** Benign
* **breast_malignant:** Malignant
---
### 4. Cervical Cancer ↪ [Reference](https://www.kaggle.com/datasets/prahladmehandiratta/cervical-cancer-largest-dataset-sipakmed)
**Imaging Modality:** Cytology / Histopathology (Pap smear / cervical slide)
* **cervix_dyk:** Dyskeratotic
* **cervix_koc:** Koilocytotic
* **cervix_mep:** Metaplastic
* **cervix_pab:** Parabasal
* **cervix_sfi:** Superficial-Intermediate
---
### 5. Kidney Cancer ↪ [Reference](https://www.kaggle.com/datasets/nazmul0087/ct-kidney-dataset-normal-cyst-tumor-and-stone)
**Imaging Modality:** Computed Tomography (CT scans)
* **kidney_normal:** Normal
* **kidney_tumor:** Tumor
---
### 6. Lung and Colon Cancer ↪ [Reference](https://www.kaggle.com/datasets/biplobdey/lung-and-colon-cancer)
**Imaging Modality:** Histopathology (Microscopy of H&E-stained slides)
* **colon_aca:** Colon Adenocarcinoma
* **colon_bnt:** Colon Benign Tissue
* **lung_aca:** Lung Adenocarcinoma
* **lung_bnt:** Lung Benign Tissue
* **lung_scc:** Lung Squamous Cell Carcinoma
---
### 7. Lymphoma ↪ [Reference](https://www.kaggle.com/datasets/andrewmvd/malignant-lymphoma-classification)
**Imaging Modality:** Histopathology / Microscopy (hematologic tissue slides)
* **lymph_cll:** Chronic Lymphocytic Leukemia
* **lymph_fl:** Follicular Lymphoma
* **lymph_mcl:** Mantle Cell Lymphoma
---
### 8. Oral Cancer ↪ [Reference](https://www.kaggle.com/datasets/ashenafifasilkebede/dataset)
**Imaging Modality:** Histopathology (H&E-stained oral cavity tissue slides)
* **oral_normal:** Normal
* **oral_scc:** Oral Squamous Cell Carcinoma
---
## 🧾 Dataset and Resource Links
* **Full Preprocessed Dataset DOI:** [Cancer Classification](https://doi.org/10.34740/kaggle/dsv/9419092)
* **Demo Video:** [https://youtu.be/GQ7QS0NIviA?si=IRA5Ncn5bzYd0wdm](https://youtu.be/GQ7QS0NIviA?si=IRA5Ncn5bzYd0wdm)
* **Test on Web:** [Multi-Cancer Classification – DAS medhub](https://sebukpor.github.io/multi-cancer-classification/)
---
## Live Demo
Try the live demo on Hugging Face:
🔗 [https://huggingface.co/spaces/Sebukpor/multi-cancer-gradcam](https://huggingface.co/spaces/Sebukpor/multi-cancer-gradcam)
*Upload sample medical images and receive instant predictions — all processed in your browser.*
---
## License
This project is licensed under the **MIT License**. See the [LICENSE](LICENSE) file for details.
---
## Contact
📧 **Email:** [divinesebukpor@gmail.com](mailto:divinesebukpor@gmail.com)
---
## Acknowledgements
We thank all dataset providers, research collaborators, and open-source communities that contributed to the datasets, deployment, and web integration.