Machine learning methods for automated classification of tumors with papillary thyroid carcinoma-like nuclei: A quantitative analysis

Böhland, Moritz; Tharun, Lars; Scherr, Tim; Mikut, Ralf; Hagenmeyer, Veit; Thompson, Lester D.�R.; Perner, Sven; Reischl, Markus

doi:10.1371/journal.pone.0257635

Cited by 22 publications

(41 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For this, we used two image tiles from lymph node dataset 5. Manual segmentation was performed by referencing all of the image channels; however, manual segmentation is biased since it relies on prior assumptions 28 , and in this case, the DAPI and CD45 channels were most heavily relied on by the experts we used. Supplementary Table 9 shows that the two expert segmentations have an average Dice coefficient of 0.6987, which is in line with prior work from 28 .…”

Section: Resultsmentioning

confidence: 99%

“…Manual segmentation was performed by referencing all of the image channels; however, manual segmentation is biased since it relies on prior assumptions 28 , and in this case, the DAPI and CD45 channels were most heavily relied on by the experts we used. Supplementary Table 9 shows that the two expert segmentations have an average Dice coefficient of 0.6987, which is in line with prior work from 28 . Results from comparing different segmentation methods with the first expert annotation, presented in Supplementary Table 7 , show that on average, RAMCES segmentation combining the top 3 (and top 2 and 4) ranked proteins improve on the average performance of the top three ranked markers (improvement of 2.1% using Jaccard and 1.3% for Dice).…”

Section: Resultsmentioning

confidence: 99%

“…This is likely because, as mentioned, the experts referenced the CD45 channel most often when manually segmenting the tiles. This highlights the challenge with obtaining ground truth since any manually-curated ground truth is likely biased by the individual performing the segmentation 28 . By combining the top three channels, RAMCES is able to overcome such biases and obtain results that are in good agreement with both expert curation and expected cell populations.…”

Section: Discussionmentioning

confidence: 99%

See 2 more Smart Citations

Membrane marker selection for segmenting single cell spatial proteomics data

et al. 2022

View full text Add to dashboard Cite

The ability to profile spatial proteomics at the single cell level enables the study of cell types, their spatial distribution, and interactions in several tissues and conditions. Current methods for cell segmentation in such studies rely on known membrane or cell boundary markers. However, for many tissues, an optimal set of markers is not known, and even within a tissue, different cell types may express different markers. Here we present RAMCES, a method that uses a convolutional neural network to learn the optimal markers for a new sample and outputs a weighted combination of the selected markers for segmentation. Testing RAMCES on several existing datasets indicates that it correctly identifies cell boundary markers, improving on methods that rely on a single marker or those that extend nuclei segmentations. Application to new spatial proteomics data demonstrates its usefulness for accurately assigning cell types based on the proteins expressed in segmented cells.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Membrane marker selection for segmenting single cell spatial proteomics data

et al. 2022

View full text Add to dashboard Cite

show abstract

“…and mild nuclear atypia (chromatin margins, nuclear membrane irregularities, etc. ), which are confused with malignant lesions; cell enlargement can also be seen locally in benign thyroid tissue ( Girolami et al, 2020 ; Böhland et al, 2021 ). Benign thyroid tissue may also have a background of fibrosis and calcification at the background level, which is common in thyroid cancer.…”

Section: Discussionmentioning

confidence: 99%

Deep Learning-Based Recognition of Different Thyroid Cancer Categories Using Whole Frozen-Slide Images

Zhu

Chen

Guo

et al. 2022

Front. Bioeng. Biotechnol.

View full text Add to dashboard Cite

Introduction: The pathological rare category of thyroid is a type of lesion with a low incidence rate and is easily misdiagnosed in clinical practice, which directly affects a patient’s treatment decision. However, it has not been adequately investigated to recognize the rare, benign, and malignant categories of thyroid using the deep learning method and recommend the rare to pathologists.Methods: We present an empirical decision tree based on the binary classification results of the patch-based UNet model to predict rare categories and recommend annotated lesion areas to be rereviewed by pathologists.Results: Applying this framework to 1,374 whole-slide images (WSIs) of frozen sections from thyroid lesions, we obtained an area under a curve of 0.946 and 0.986 for the test datasets with and without WSIs, respectively, of rare types. However, the recognition error rate for the rare categories was significantly higher than that for the benign and malignant categories (p < 0.00001). For rare WSIs, the addition of the empirical decision tree obtained a recall rate and precision of 0.882 and 0.498, respectively; the rare types (only 33.4% of all WSIs) were further recommended to be rereviewed by pathologists. Additionally, we demonstrated that the performance of our framework was comparable to that of pathologists in clinical practice for the predicted benign and malignant sections.Conclusion: Our study provides a baseline for the recommendation of the uncertain predicted rare category to pathologists, offering potential feasibility for the improvement of pathologists’ work efficiency.

show abstract

“…At first, we perform extensive research of existing segmentation approach [24]- [29], study their methodologies and analyze their shortcoming. Further, we design and develop improvised convolution auto encoder aka (ICAE) for thyroid gland segmentation; ICAE not only helps in segmenting and finding return on investment (ROI) but also enhances the image.…”

Section: Introductionmentioning

confidence: 99%

Improvised convolutional auto encoder for thyroid nodule image enhancement and segmentation

Gunjali¹,

Hansavath

2022

IJEECS

View full text Add to dashboard Cite

Thyroid <span>ultrasonography and thermography are a widely used clinical technique for nodule diagnosis in thyroid regions. However, it remains difficult to detect and recognize the nodules due to low contrast, high noise, and diverse appearance of nodules. To alleviate doctors’ tremendous labor in the diagnosis procedure, we advocate a machine learning approach to the detection and recognition tasks in this paper. Moreover, this research mainly focuses on segmenting the image and finding the probable region. In this research work an improvised convolutional auto encoder (ICAE) is introduced for segmenting the image and finding the probable region of thyroid gland and it enhances image. ICAE comprises various layer and mechanism, each having their own task. Apart from the traditional approach, skip connection is applied for the image enhancement and dual frame is introduced for better feature extraction. Further optimization technique is used for increasing the learning rate. ICAE is evaluated considering digital database thyroid image (DDTI) dataset with performance metrics like accuracy, true positive rate, false positive rate, dice coefficient and similarity index (SI); also, comparative analysis is carried out with various existing model and proposed model simply outperforms the existing model.</span>

show abstract

Machine learning methods for automated classification of tumors with papillary thyroid carcinoma-like nuclei: A quantitative analysis

Cited by 22 publications

References 24 publications

Membrane marker selection for segmenting single cell spatial proteomics data

Membrane marker selection for segmenting single cell spatial proteomics data

Deep Learning-Based Recognition of Different Thyroid Cancer Categories Using Whole Frozen-Slide Images

Improvised convolutional auto encoder for thyroid nodule image enhancement and segmentation

Contact Info

Product

Resources

About