🧬 ALLNet

Acute Lymphoblastic Leukemia Detection using Lightweight Convolutional Networks

Source code for the IEEE CIVEMSA 2022 paper
ALLNet: Acute Lymphoblastic Leukemia Detection using lightweight convolutional networks

---

🔬 Overview

ALLNet is a lightweight deep-learning pipeline for Acute Lymphoblastic Leukemia (ALL) detection from blood-cell images.

The repository implements a two-stage transfer-learning workflow:

1. Pre-training a lightweight convolutional network on the Atlas of Digital Pathology (ADP) dataset.
2. Fine-tuning the pre-trained model on the ALL-IDB dataset for ALL detection.

The approach is designed to exploit histopathological pre-training and adapt the learned visual representations to leukemic-cell classification.

---

🧠 Method at a Glance

ADP histopathology patches
        │
        ▼
LBCNN / lightweight CNN pre-training
        │
        ▼
Pre-trained models
        │
        ▼
Fine-tuning on ALL-IDB blood-cell images
        │
        ▼
Acute Lymphoblastic Leukemia detection

---

✨ Key Features

• 🧬 ALL detection from microscopic blood-cell images
• ⚡ Lightweight convolutional architecture based on LBCNN components
• 🔁 Transfer-learning workflow from ADP to ALL-IDB
• 🧪 Two-stage experimental pipeline: pre-training and fine-tuning
• 🐍 Python / PyTorch implementation
• 📄 Reproducible code accompanying the IEEE CIVEMSA 2022 paper

---

📁 Repository Structure

ALLNet/
│
├── 0_PyTorch_ADP_HistoNet_LBCNN/
│   ├── pytorch_adp_histonet.py
│   └── lbcnn_pytorch/
│       └── requirements.txt
│
├── 1_PyTorch_ADP_HistoNet_LBCNN_fineTune_ALL_IDB/
│   ├── pytorch_adp_histonet_finetune_all.py
│   ├── pretrained_nets/
│   └── lbcnn_pytorch/
│       └── requirements.txt
│
├── LICENSE
└── README.md

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/AngeloUNIMI/ALLNet.git
cd ALLNet

2. Create a Python environment

Using Conda:

conda create -n allnet python=3.7
conda activate allnet

Install the required packages from the corresponding experiment folder:

cd 0_PyTorch_ADP_HistoNet_LBCNN/lbcnn_pytorch
pip install -r requirements.txt

The repository dependency graph reports the main framework versions as:

pytorch == 1.4.0
torchvision == 0.5.0
tqdm

---

🗂 Dataset Preparation

ADP dataset for pre-training

Download the Atlas of Digital Pathology (ADP) dataset from:

https://www.dsp.utoronto.ca/projects/ADP/

Expected files and folders:

ADP/img_res_1um_bicubic/
ADP/ADP_EncodedLabels_Release1_Flat.csv

The ADP images should be split into patches following the dataset instructions.

ALL-IDB dataset for fine-tuning

Download the Acute Lymphoblastic Leukemia Image Database for Image Processing (ALL-IDB) from:

https://homes.di.unimi.it/scotti/all/

Expected folder:

ALL_IDB2/

---

🏃 Running the Pipeline

Step 1 — Pre-train on ADP

cd 0_PyTorch_ADP_HistoNet_LBCNN
python pytorch_adp_histonet.py

This step trains the lightweight CNN on the ADP histopathology database.

Step 2 — Copy pre-trained models

Copy the generated models into:

1_PyTorch_ADP_HistoNet_LBCNN_fineTune_ALL_IDB/pretrained_nets/

Some pre-trained models may already be included for convenience.

Step 3 — Fine-tune on ALL-IDB

cd 1_PyTorch_ADP_HistoNet_LBCNN_fineTune_ALL_IDB
python pytorch_adp_histonet_finetune_all.py

This step fine-tunes the network for Acute Lymphoblastic Leukemia detection on ALL-IDB.

---

📊 Expected Workflow Outputs

Depending on the script configuration, the pipeline can produce:

Stage	Output
ADP pre-training	Trained lightweight CNN models
Model transfer	Pre-trained checkpoints for fine-tuning
ALL-IDB fine-tuning	ALL classification model
Evaluation	Classification performance on ALL-IDB

---

📚 Datasets

Dataset	Purpose	Link
ADP — Atlas of Digital Pathology	Histopathology pre-training	https://www.dsp.utoronto.ca/projects/ADP/
ALL-IDB	Acute Lymphoblastic Leukemia detection	https://homes.di.unimi.it/scotti/all/

ADP Reference

Mahdi S. Hosseini, Lyndon Chan, Gabriel Tse, Michael Tang, Jun Deng, Sajad Norouzi, Corwyn Rowsell, Konstantinos N. Plataniotis, and Savvas Damaskinos,
“Atlas of Digital Pathology: A Generalized Hierarchical Histological Tissue Type-Annotated Database for Deep Learning,”
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 11747–11756.

ALL-IDB Reference

R. Donida Labati, V. Piuri, and F. Scotti,
“ALL-IDB: the acute lymphoblastic leukemia image database for image processing,”
Proc. of the 2011 IEEE International Conference on Image Processing (ICIP 2011), Brussels, Belgium, pp. 2045–2048, September 11–14, 2011.
DOI: 10.1109/ICIP.2011.6115881

---

📖 Paper

If you use this code, please cite:

@InProceedings{civemsa22_all,
  author    = {A. Genovese},
  booktitle = {Proc. of the 2022 IEEE Int. Conf. on Computational Intelligence and Virtual Environments for Measurement Systems and Applications (CIVEMSA 2022)},
  title     = {ALLNet: Acute Lymphoblastic Leukemia Detection using lightweight convolutional networks},
  address   = {Chemnitz, Germany},
  pages     = {1--6},
  month     = {June},
  day       = {15--17},
  year      = {2022},
  note      = {Accepted}
}

Paper:

https://ieeexplore.ieee.org/document/9853691

Project page:

https://iebil.di.unimi.it/cnnALL/index.htm

---

👤 Author

Angelo Genovese
Department of Computer Science
Università degli Studi di Milano, Italy

---

📄 License

This project is released under the GNU General Public License v3.0.

See the LICENSE file for details.