Whole-slide Multiple Instance Learning for Predicting Axillary Lymph Node Metastases

This repo is the official implementation of the paper "Whole-slide Multiple Instance Learning for Predicting Axillary Lymph Node Metastases".

Abstract

Breast cancer is a major concern for women's health globally, with axillary lymph node (ALN) metastasis identification being critical for prognosis evaluation and treatment guidance. This paper presents a deep learning (DL) classification pipeline for quantifying clinical information from digital core-needle biopsy (CNB) images, with fewer steps than existing methods. A publicly available dataset of 1058 patients was used to evaluate the performance of different baseline state-of-the-art (SOTA) DL models in classifying ALN metastatic status based on CNB images. An extensive ablation study of various data augmentation techniques was also conducted. Finally, the manual tumor segmentation and annotation step performed by the pathologists was assessed. Our proposed training scheme outperformed SOTA by 3.73%.

Setup

Clone this repo

git clone https://github.com/glejdis/MICCAI_2023.git

Environment

Create environment and install dependencies.

conda create -n DLCNBC python=3.9.12 -y
conda activate DLCNBC
pip install -r code/requirements.txt

Dataset

For your convenience, we have provided preprocessed clinical data in code/dataset. The processed WSI patches can be downloaded from here and unzip them by the following scripts:

cd code/dataset
# download paper_patches.zip
unzip paper_patches.zip

Illustration of exemplary patches from each class of the BCNB dataset:

Training

To train our models run the following:

experiment_index:

0. N0 vs N+(>0)
1. N0 vs N+(1-2) vs N+(>2)

To run any experiment of the DLCNB with the clinical data, you can do as this:

cd code
bash run.sh ${experiment_index}

To run any experiment of the DLCNB without the clinical data, you can do as this:

cd code
bash run_no_clinical.sh ${experiment_index}

To run any experiment of the DLCNB with the clinical data and data augmentation strategies, you can do as this:

cd code
bash run_further_data_aug.sh ${experiment_index}

Furthermore, if you want to try other settings, please see train.py for more details.

Results

Some of the best results obtained from our experiemnts are given in folders plots, plots_no_clinical, plots_no_Segmentation and logs.

The overall performance results in terms of AUC score of top 10 data augmentation techniques in the prediction of ALN status: DL-CNB+C model (N0 vs. N(+)) with VGG-16 with BN is displayed below. The labels on the right represent the AUC score on the test cohort, and the methods are sorted in descending order from the highest to the lowest AUC score.

Image below displays the confusion matrix (left) and the ROC curves (right) of DL-CNB+C model with VGG-13 with BN for feature extraction in predicting N0 and N(+) classes in the test cohort (0: N0, 1: N(+)).

Image below shows the confusion matrix (left) and the ROC curves (right) of DL-CNB+C model with VGG-13 BN for feature extraction without tumor segmentation in predicting N0 and N(+) classes in the test cohort (0: N0, 1: N(+)).

Image below shows the confusion matrix and the comparison of ROC curves between different classes when using DL-CNB+C model with VGG-16 BN for feature extraction in predicting N0, N+(1-2) and N+(>2) classes (0: N0, 1: N+(1-2), 2: N+(>2)).