PDBL: Improving Histopathological Tissue Classification With Plug-and-Play Pyramidal Deep-Broad Learning

Lin, Jiatai; Han, Guoqiang; Pan, Xipeng; Liu, Zaiyi; Chen, Hao; Li, Danyi; Jia, Xiping; Shi, Zhenwei; Wang, Zhizhen; Cui, Yanfen; Li, Haiming; Liang, Changhong; Li, Liang; Wang, Ying; Han, Chu

doi:10.1109/tmi.2022.3161787

Cited by 36 publications

(7 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, the prediction accuracy of deep learning model for recognition of TME components should be improved, to ensure the calculated TME features are closer to the real situation. Some new models such as PDBL ( 31 ), CRCCN-Net ( 32 ) and Vision Transformer ( 33 ) are worthy of being using, since they have achieved accuracy of more than 96% in Kather’s dataset. Third, nearly half TME signature associated genes failed to be identified by GO and KEGG databases, which may affect the comprehensiveness of functional annotations for TME signature.…”

Section: Discussionmentioning

confidence: 99%

Prognostic prediction based on histopathologic features of tumor microenvironment in colorectal cancer

2023

View full text Add to dashboard Cite

PurposeTo automatically quantify colorectal tumor microenvironment (TME) in hematoxylin and eosin stained whole slide images (WSIs), and to develop a TME signature for prognostic prediction in colorectal cancer (CRC).MethodsA deep learning model based on VGG19 architecture and transfer learning strategy was trained to recognize nine different tissue types in whole slide images of patients with CRC. Seven of the nine tissue types were defined as TME components besides background and debris. Then 13 TME features were calculated based on the areas of TME components. A total of 562 patients with gene expression data, survival information and WSIs were collected from The Cancer Genome Atlas project for further analysis. A TME signature for prognostic prediction was developed and validated using Cox regression method. A prognostic prediction model combined the TME signature and clinical variables was also established. At last, gene-set enrichment analysis was performed to identify the significant TME signature associated pathways by querying Gene Ontology database and Kyoto Encyclopedia of Genes and Genomes database.ResultsThe deep learning model achieved an accuracy of 94.2% for tissue type recognition. The developed TME signature was found significantly associated to progression-free survival. The clinical combined model achieved a concordance index of 0.714. Gene-set enrichment analysis revealed the TME signature associated genes were enriched in neuroactive ligand-receptor interaction pathway.ConclusionThe TME signature was proved to be a prognostic factor and the associated biologic pathways would be beneficial to a better understanding of TME in CRC patients.

show abstract

Section: Discussionmentioning

confidence: 99%

Prognostic prediction based on histopathologic features of tumor microenvironment in colorectal cancer

2023

View full text Add to dashboard Cite

show abstract

“…They used patch-level labels for the estimation of pixel-level labels using weak supervision for tissue semantic segmentation. Li et al proposed a pyramidal deep broad learning method for tissue classification [29].…”

Section: A Tissue Phenotyping Methodsmentioning

confidence: 99%

Knowledge Distillation in Histology Landscape by Multi-Layer Features Supervision

Javed

Mahmood

Qaiser

et al. 2023

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

Automatic tissue classification is a fundamental task in computational pathology for profiling tumor micro-environments. Deep learning has advanced tissue classification performance at the cost of significant computational power. Shallow networks have also been end-toend trained using direct supervision however their performance degrades because of the lack of capturing robust tissue heterogeneity. Knowledge distillation has recently been employed to improve the performance of the shallow networks used as student networks by using additional supervision from deep neural networks used as teacher networks. In the current work, we propose a novel knowledge distillation algorithm to improve the performance of shallow networks for tissue phenotyping in histology images. For this purpose, we propose multi-layer feature distillation such that a single layer in the student network gets supervision from multiple teacher layers. In the proposed algorithm, the size of the feature map of two layers is matched by using a learnable multi-layer perceptron. The distance between the feature maps of the two layers is then minimized during the training of the student network. The overall objective function is computed by summation of the loss over multiple layers combination weighted with a learnable attention-based parameter. The proposed algorithm is named as Knowledge Distillation for Tissue Phenotyping (KDTP). Experiments are performed on five different publicly available histology image classification datasets using several teacher-student network combinations within the KDTP algorithm. Our results demonstrate a significant performance increase in the student networks by using the proposed KDTP algorithm compared to direct supervisionbased training methods.

show abstract

“…After discussion with pathologists from Shenzhen Third People's Hospital and investigating related papers [44,45], We can see that the FOV of the pathologist inspecting at the WSI under the microscope at different resolutions is relatively fixed. Even if the FOV of the pathologist changes at different magnifications, the pathologist usually observes the same area near the center of the field.…”

Section: A Multi-magnification Histopathological Imagementioning

confidence: 98%

Deep Multi-Magnification Similarity Learning for Histopathological Image Classification

Diao

Luo

Hou

et al. 2023

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

Precise classification of histopathological images is crucial to computer-aided diagnosis in clinical practice. Magnification-based learning networks have attracted considerable attention for their ability to improve performance in histopathological classification. However, the fusion of pyramids of histopathological images at different magnifications is an under-explored area. In this paper, we proposed a novel deep multi-magnification similarity learning (DSML) approach that can be useful for the interpretation of multi-magnification learning framework and easy to visualize feature representation from low-dimension (e.g., cell-level) to high-dimension (e.g., tissue-level), which has overcome the difficulty of understanding cross-magnification information propagation. It uses a similarity cross entropy loss function designation to simultaneously learn the similarity of the information among cross-magnifications. In order to verify the effectiveness of DMSL, experiments with different network backbones and different magnification combinations were designed, and its ability to interpret was also investigated through visualization. Our experiments were performed on two different histopathological datasets: a clinical nasopharyngeal carcinoma and a public breast cancer BCSS2021 dataset.The results show that our method achieved outstanding performance in classification with a higher value of area under curve, accuracy, and F-score than other comparable methods. Moreover, the reasons behind multimagnification effectiveness were discussed.

show abstract

PDBL: Improving Histopathological Tissue Classification With Plug-and-Play Pyramidal Deep-Broad Learning

Cited by 36 publications

References 40 publications

Prognostic prediction based on histopathologic features of tumor microenvironment in colorectal cancer

Prognostic prediction based on histopathologic features of tumor microenvironment in colorectal cancer

Knowledge Distillation in Histology Landscape by Multi-Layer Features Supervision

Deep Multi-Magnification Similarity Learning for Histopathological Image Classification

Contact Info

Product

Resources

About