HID-CON: Weakly Supervised Intrahepatic Cholangiocarcinoma Subtype Classification of Whole Slide Images using Contrastive Hidden Class Detection

By Jing Wei Tan, Kyoungbun Lee and Won-Ki Jeong

Paper

Abstract

Purpose: Biliary tract cancer, also known as intrahepatic cholangiocarcinoma (IHCC), is a rare disease that shows no clear symptoms during its early stage, but its prognosis depends highly on the cancer subtype. Hence, an accurate cancer subtype classification model is necessary to provide better treatment plans to patients and to reduce mortality. However, annotating histopathology images at the pixel or patch level is time-consuming and labor-intensive for giga-pixel whole slide images. To address this problem, we propose a weakly supervised method for classifying IHCC subtypes using only image-level labels. Approach: The core idea of the proposed method is to detect regions (i.e., subimages or patches) commonly included in all subtypes, which we name the “hidden class,” and to remove them via iterative application of contrastive loss and label smoothing. Doing so will enable us to obtain only patches that faithfully represent each subtype, which are then used to train the image-level classification model by multiple instance learning (MIL). Results: Our method outperforms the state-of-the-art weakly supervised learning methods ABMIL, TransMIL, and DTFD-MIL by ∼17%, 18%, and 8%, respectively, and achieves performance comparable to that of supervised methods. Conclusion: The introduction of a hidden class to represent patches commonly found across all subtypes enhances the accuracy of IHCC classification and addresses the weak labeling problem in histopathology images.

Framework

Tensorflow 2

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).

Warm-up Stage Training

1. Training warm-up model

 python train_warmup.py --gpus [gpu's index] --epochs [number of epochs] --batch_size [number of batch] --training_set [training patch list in csv] --validation_set [training patch list in csv] --label [label list in csv] --ckpt_dir [checkpoint directory] 

2. Deploy Warm-up model

python patch_label_prediction.py --training_set [Patch list for validation set] --validation_set [Patch list for validation set] --ckpt_dir [Checkpoint directory] --ckpt [checkpoint] --saving_dir [Directory to save the predicted label]  --task warm-up --batch_size [batch size]

3. Topk patch filtering

python topk.py --saving_dir [folder directory to be saved] --saving_folder [saving folder name]  --set_label [patch list] --set [tr/val]

HID-CON Training

1. Training HID-CON model

python train_hidcon.py --gpus [gpu index] --epochs [epoch number] --batch_size [batch size] --main_dir [directory] --label [directory for label] --ckpt_dir [directory for saving checkpoint] --prev_ckpt [previous checkpoint name] --prev_prev_ckpt [previous 2 iteration checkpoint name] --iter_num [iteration number]

2. Deploy Warm-up model

python patch_label_prediction.py --training_set [Patch list for validation set] --validation_set [Patch list for validation set] --ckpt_dir [Checkpoint directory] --ckpt [checkpoint] --saving_dir [Directory to save the predicted label]  --task warm-up --batch_size [batch size]

3. Topk patch filtering for next iteration training

python topk.py --saving_dir [folder directory to be saved] --saving_folder [saving folder name]  --set_label [patch list] --set [tr/val]

4. Hidden class filtering for downstraem task

python hidden_filtering.py --main_dir [directory] --saving_folder [folder taht saved the patch label prediction]

Bag Classification with MIL

1. Bag list generation （Example of bag_list.csv).

python bag_list.py --main_dir [directoty] --patch_list [patch_list after filtering put hidden class patches] --num_bag 50 --num_patch 100

2. Bag feature generation (Example of split_file.csv.)

 python bag_feature_generation.py  --main_dir [directory] --bag_list [sample_images_xxxx_50b100p] --rawpatch_dir [the directory that stored raw patches] --split_file [csv file that stores train or test label]

3. Bag classification

python bag_classification_github.py  --main_dir [directory] --bag_folder [folder that stored bag features] --label [label in csv] --batch_size [batch size]  --lr [learning rate] --dc [decay weight] --epochs [number of epoch] --iter [iteration number] --save_ckpt_dir [directory for saving checkpoint]

Inference

The 'best_aver.npy' should be first downlaoded from this page first.

 python inference.py --model_path [.../best.hdf5] --input_file [../xx.csv] --save_path [/xxx/xxx] --aver_path [.../best_aver.npy]

The checkpoint can be downloaded from here.

Citation

Please cite us if you use our work.

Tan, Jing Wei, Kyoungbun Lee, and Won-Ki Jeong. "HID-CON: weakly supervised intrahepatic cholangiocarcinoma subtype classification of whole slide images using contrastive hidden class detection." Journal of Medical Imaging 12, no. 6 (2025): 061402-061402.