[MICCAI 2023] Histopathology Image Classification using Deep Manifold Contrastive Learning

By Jing Wei Tan and Won-Ki Jeong

[Paper]

This is the official implementation code.

Abstract

Contrastive learning has gained popularity due to its robustness with good feature representation performance. However, cosine distance, the commonly used similarity metric in contrastive learning, is not well suited to represent the distance between two data points, especially on a nonlinear feature manifold. Inspired by manifold learning, we propose a novel extension of contrastive learning that leverages geodesic distance between features as a similarity metric for histopathology whole slide image classification. To reduce the computational overhead in manifold learning, we propose geodesic-distance-based feature clustering for efficient contrastive loss evaluation using prototypes without time-consuming pairwise feature similarity comparison. The efficacy of the proposed method is evaluated on two real-world histopathology image datasets. Results demonstrate that our method outperforms state-of-the-art cosine-distance-based contrastive learning methods.

Framework

Tensorflow 2

Dataset

The hepatocellular carcinomas (HCCs) dataset can be downloaded from Pathology AI Platform.

Preprocessing

1. Download the raw WSI data.
2. Prepare the patches.
3. Store all the patches directory in a .csv file (refer patch_list.csv).

Deep Manifold Embedding Learning

Manifold encoder training.

python manifold.py --tr_patch_list [CSV path] --val_patch_list [CSV path] --label_file [CSV path] --save_dir [folder path] --num_class 2 --num_NN 5 --num_cluster 10 --save_model_dir [folder path]

Bag formation

1. Generate bag list in bag_list.csv.

python bag_list_generation.py --tr_patch_list [CSV path] --val_patch_list [CSV path] --te_patch_list [CSV path] --save_dir [folder path] --num_bag 50 --num_patchPerbag 100 

2. Feature extration based on bag_list.csv. Example of split_file.csv is in split_file.csv.

python feature_extraction.py --bag_list_dir  [CSV path]  --split_file [CSV path] --ckpt_dir [checkpoint path] --save_dir [folder path] --src_dir [folder_path]

MIL training

Train simple MIL for classification.

 python mil.py --feat_dir  [folder path]  --label_file [CSV path] --num_class 2 --save_model_dir [CSV path]

Citation

Please cite us if you use our work.

Tan, Jing Wei, and Won-Ki Jeong. "Histopathology image classification using deep manifold contrastive learning." In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 683-692. Cham: Springer Nature Switzerland, 2023.