A Generalized Deep Learning Framework for Whole-Slide Image Segmentation and Analysis

Khened, Mahendra; Kori, Avinash; Rajkumar, Haran; Srinivasan, Balaji; Krishnamurthi, Ganapathy

doi:10.48550/arxiv.2001.00258

Cited by 2 publications

(6 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such unsatisfactory result makes us believe thousands of training whole slide images is necessary for image-level pipeline to converge. Pixel-level [53] setting reaches 90.90%, 5th in leaderboard, whose generalization has been validated in various challenges. Our mixed supervision learning is the 92.43%, 4th, which is 1.53% more than pixellevel baseline.…”

Section: E Evaluation On the Camelyon17mentioning

confidence: 76%

“…It is K = 1 in the released code [14]. Pixel-level experiment is conducted based on source code of Mahendra Khened et al [53]. This generalized pathology processing framework is the 5th in Camelyon17 Challenge 1 [16], 4th in DigestPath2019 2 [15] and 3rd in PAIP challenge 3 .…”

Section: A Experimental Settingmentioning

confidence: 99%

“…The reason for this phenomenon is that with image-level Based on code of Mahendra Khened et al [53], 200 pixellevel fine-grained labeled big patches are used to train the initial patch segmentation model. Then, top patches are selected for whole slide images classifier training, without imagelevel labels guided pixel-level pseudo labels generation.…”

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

“…2) Results: The same as the gastric cancer dataset, models are trained on image-level [14], pixel-level [53] and mixed setting. Image-level [14] experiment on cervical cytology failed to converge for extremely tiny positive cells ratio.…”

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

“…Image-level [14] experiment on cervical cytology failed to converge for extremely tiny positive cells ratio. In pixel-level training [53], initial labels are provided on center B. Specificity 81% is achieved on center B but drops to 72% on center A due to domain gap. To solve domain gap problem the direct solution is to train data from that source, even on noisy pseudo labeled data.…”

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

See 4 more Smart Citations

Hybrid Supervision Learning for Pathology Whole Slide Image Classification

Li¹,

Chen²,

Huang³

et al. 2021

Preprint

View full text Add to dashboard Cite

Weak supervision learning on classification labels has demonstrated high performance in various tasks. When a few pixel-level fine annotations are also affordable, it is natural to leverage both of the pixel-level (e.g., segmentation) and image level (e.g., classification) annotation to further improve the performance. In computational pathology, however, such weak or mixed supervision learning is still a challenging task, since the high resolution of whole slide images makes it unattainable to perform endto-end training of classification models. An alternative approach is to analyze such data by patch-base model training, i.e., using self-supervised learning to generate pixellevel pseudo labels for patches. However, such methods usually have model drifting issues, i.e., hard to converge, because the noise accumulates during the self-training process. To handle those problems, we propose a mixed supervision learning framework for super high-resolution images to effectively utilize their various labels (e.g., sufficient image-level coarse annotations and a few pixel-level fine labels). During the patch training stage, this framework can make use of coarse image-level labels to refine selfsupervised learning and generate high-quality pixel-level pseudo labels. A comprehensive strategy is proposed to suppress pixel-level false positives and false negatives. Three real-world datasets with very large number of images (i.e., more than 10,000 whole slide images) and various types of labels are used to evaluate the effectiveness of mixed supervision learning. We reduced the false positive rate by around one third compared to state of the art while retaining 100% sensitivity, in the task of image-level classification.

show abstract

Section: E Evaluation On the Camelyon17mentioning

confidence: 76%

Section: A Experimental Settingmentioning

confidence: 99%

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

Section: Evaluation On the Gastric Cancer Datasetmentioning

confidence: 99%

See 3 more Smart Citations

Hybrid Supervision Learning for Pathology Whole Slide Image Classification

Li¹,

Chen²,

Huang³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

Colorectal Cancer Segmentation using Atrous Convolution and Residual Enhanced UNet

Shah¹,

Gupta²,

Lodaya³

et al. 2021

Preprint

View full text Add to dashboard Cite

Colorectal cancer is a leading cause of death worldwide. However, early diagnosis dramatically increases the chances of survival, for which it is crucial to identify the tumor in the body. Since its imaging uses high-resolution techniques, annotating the tumor is time-consuming and requires particular expertise. Lately, methods built upon Convolutional Neural Networks(CNNs) have proven to be at par, if not better in many biomedical segmentation tasks. For the task at hand, we propose another CNN-based approach, which uses atrous convolutions and residual connections besides the conventional filters. The training and inference were made using an efficient patch-based approach, which significantly reduced unnecessary computations. The proposed AtResUNet was trained on the DigestPath 2019 Challenge dataset for colorectal cancer segmentation with results having a Dice Coefficient of 0.748. Its ensemble, with its simpler version, achieved a Dice Coefficient of 0.753. D. Gupta, R. Lodaya, U. Baid contributed equally to this article

show abstract

A Generalized Deep Learning Framework for Whole-Slide Image Segmentation and Analysis

Cited by 2 publications

References 41 publications

Hybrid Supervision Learning for Pathology Whole Slide Image Classification

Hybrid Supervision Learning for Pathology Whole Slide Image Classification

Colorectal Cancer Segmentation using Atrous Convolution and Residual Enhanced UNet

Contact Info

Product

Resources

About