All papers that have not been peer-reviewed will not appear here, including preprints. You can access my all of papers at 🔗Google Scholar.
2024
SeqFRT: Towards Effective Adaption of Foundation Model via Sequence Feature Reconstruction in Computational Pathology
Chengfei Cai, Jun Li, Mingxin Liu, Yiping Jiao, Jun Xu†(† corresponding author)
2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 2024 ConferenceOral
In this paper, we present an innovative weakly-supervised sequence feature optimization method to solve the problem of sub-optimal feature extraction by the foundation model in the traditional MIL paradigm. The proposed SeqFRT leverages a sequence position optimization strategy to exploit the inherent valuable information embedded within the long pathological feature sequences and a sequence sparsity enhancement technique to highly enhance the ability to discriminate and extract the latent representations instead of redundant information, leading to preserving essential information for reconstructing input pathological sequence representation which is crucial for downstream tasks in computational pathology.
Edge and dense attention U-net for atrial scar segmentation in LGE-MRI
Gaoyuan Li, Mingxin Liu, Jun Lu, Jiquan Ma†(† corresponding author)
Biomedical Physics & Engineering Express 2024 Journal
We introduce a dual branch network, incorporating edge attention, and deep supervision strategy. Edge attention is introduced to fully utilize the spatial relationship between the scar and the atrium. Besides, dense attention is embedded in the bottom layer to solve feature disappearance. At the same time, deep supervision accelerates the convergence of the model and improves segmentation accuracy.
Exploiting Geometric Features via Hierarchical Graph Pyramid Transformer for Cancer Diagnosis Using Histopathological Images
Mingxin Liu, Yunzan Liu, Pengbo Xu, Hui Cui, Jing Ke, Jiquan Ma†(† corresponding author)
IEEE Transactions on Medical Imaging 2024 Journal
This study proposed HGPT, a novel framework that jointly considers geometric and global representation for cancer diagnosis in histopathological images. HGPT leverages a multi-head graph aggregator to aggregate the geometric representation from pathological morphological features, and a locality feature enhancement block to highly enhance the 2D local feature perception in vision transformers, leading to improved performance on histopathological image classification. Extensive experiments on Kather-5K, MHIST, NCT-CRC-HE, and GasHisSDB four public datasets demonstrate the advantages of the proposed HGPT over bleeding-edge approaches in improving cancer diagnosis performance.
Unleashing the Infinity Power of Geometry：A Novel Geometry-Aware Transformer (GOAT) for Whole Slide Histopathology Image Analysis
Mingxin Liu, Yunzan Liu, Pengbo Xu, Jiquan Ma†(† corresponding author)
2024 IEEE International Symposium on Biomedical Imaging (ISBI) 2024 ConferenceOral
We proposed a novel weakly-supervised framework, Geometry-Aware Transformer (GOAT), in which we urge the model to pay attention to the geometric characteristics within the tumor microenvironment which often serve as potent indicators. In addition, a context-aware attention mechanism is designed to extract and enhance the morphological features within WSIs. Extensive experimental results demonstrated that the proposed method is capable of consistently reaching superior classification outcomes for gigapixel whole slide images.
2023
MGCT：Mutual-Guided Cross-Modality Transformer for Survival Outcome Prediction using Integrative Histopathology-Genomic Features
Mingxin Liu, Yunzan Liu, Hui Cui, Chunquan Li†, Jiquan Ma†(† corresponding author)
2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 2023 ConferenceOral
We propose the Mutual-Guided Cross-Modality Transformer (MGCT), a weakly-supervised, attention-based multimodal learning framework that can combine histology features and genomic features to model the genotype-phenotype interactions within the tumor microenvironment. Extensive experimental results on five benchmark datasets consistently emphasize that MGCT outperforms the state-of-the-art (SOTA) methods.