Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks

Cireşan, Dan; Giusti, Alessandro; Gambardella, Luca Maria; Schmidhuber, Juergen

doi:10.1007/978-3-642-40763-5_51

Cited by 1,070 publications

(780 citation statements)

References 16 publications

(17 reference statements)

Supporting

Mentioning

740

Contrasting

Unclassified

Order By: Relevance

“…The downsides of this approach are that it does not use all the information content of the images and may bias towards rare classes. Hierarchical training [5,21,20] and retraining [9] have been proposed as alternative strategies but they can be prone to overfitting and sensitive to the state of the initial classifiers [10]. Recent training methods for FCNs resorted to loss functions based on sample re-weighting [2,10,12,16,18], where lesion regions, for example, are given more importance than non-lesion regions during training.…”

Section: Introductionmentioning

confidence: 99%

Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks

Salehi

Erdoğmuş

Gholipour

2017

Lecture Notes in Computer Science

706

380

View full text Add to dashboard Cite

Abstract. Fully convolutional deep neural networks carry out excellent potential for fast and accurate image segmentation. One of the main challenges in training these networks is data imbalance, which is particularly problematic in medical imaging applications such as lesion segmentation where the number of lesion voxels is often much lower than the number of non-lesion voxels. Training with unbalanced data can lead to predictions that are severely biased towards high precision but low recall (sensitivity), which is undesired especially in medical applications where false negatives are much less tolerable than false positives. Several methods have been proposed to deal with this problem including balanced sampling, two step training, sample re-weighting, and similarity loss functions. In this paper, we propose a generalized loss function based on the Tversky index to address the issue of data imbalance and achieve much better trade-off between precision and recall in training 3D fully convolutional deep neural networks. Experimental results in multiple sclerosis lesion segmentation on magnetic resonance images show improved F2 score, Dice coefficient, and the area under the precision-recall curve in test data. Based on these results we suggest Tversky loss function as a generalized framework to effectively train deep neural networks.

show abstract

Section: Introductionmentioning

confidence: 99%

Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks

Salehi

Erdoğmuş

Gholipour

2017

Lecture Notes in Computer Science

706

380

View full text Add to dashboard Cite

show abstract

“…Various deep learning architectures (Bengio 2009;Schmidhuber 2015), like convolutional deep neural networks (LeCun et al 1989), deep belief networks ) and recurrent neural networks (Goller and Kuchler 1996) have been applied to a lot of fields such as image recognition, speech recognition, natural language processing and bioinformatics where they have produced state-of-the-art results on various tasks (Goller and Kuchler 1996;Deng et al 2013;Deng and Yu 2014;Cireşan et al 2013;Mesnil et al 2015;Krizhevsky et al 2012;Chicco et al 2014). It becomes one of the main streams that push the development of artificial intelligence in the whole world.…”

Section: Hierarchical Structuralism: a New Mechanism For Artificial Imentioning

confidence: 99%

DGCC: data-driven granular cognitive computing

Wang

2017

Granul. Comput.

View full text Add to dashboard Cite

2016 is the 60 anniversary of the birth of artificial intelligence (AI). In the past 60 years, many AI theoretical models have been proposed and great achievements have been accomplished. Most intelligent computing models are inspired by various human/ natural/social intelligence mechanisms. Three main schools of artificial intelligence are formed, that is, symbolism, connectionism and behaviorism. Artificial intelligence is intelligence exhibited by machines. It is applied when a machine mimics cognitive functions that humans associate with other human minds. Achievements of cognitive science could give much inspiration to AI. Cognitive computing is one of the core fields of artificial intelligence. It aims to develop a coherent, unified, universal mechanism inspired by human mind's capabilities. It is one of the most critical tasks for artificial intelligence researchers to develop advanced cognitive computing models. Cognitive computing is the third and most transformational phase in computing's evolution, after the two distinct eras of computing-the tabulating era and the programming era. Inspired by human's granularity thinking, problem solving mechanism and the cognition law of ''global precedence'', a new powerful cognitive computing model, data-driven granular cognitive computing (DGCC), is proposed in this paper. It takes data as a special kind of knowledge expressed in the lowest granularity level of a multiple granularity space. It integrates two contradictory mechanisms, namely, the human's cognition mechanism of ''global precedence'' which is a cognition process of ''from coarser to finer'' and the information processing mechanism of machine learning systems which is ''from finer to coarser'', in a multiple granularity space. It is also based on the idea of data-driven. The research issues of DGCC to be further addressed are discussed. Based on DGCC, deep learning is neither classified into symbolism, nor connectionism. It is taken as a combination of symbolism and connectionism, and named hierarchical structuralism in this paper. The HD 3 characteristics (hierarchical, distributed, data-driven, and dynamical) of the hierarchical structuralism are analyzed. DGCC provides a granular cognitive computing framework for efficient knowledge discovery from big data.

show abstract

“…Although these pixel-wise textural and statistical features have been proved to be effective imaging attributes for mitosis detection, the discrimination power of these features can be degraded by artifacts present in the image due to slide preparation and acquisition. Further, Wang et al [19], Ciresan et al [20], Malon and Cosatto [21], and Albarqouni et al [22] used convolutional neural network (CNN) to detect mitosis in breast histology images. Despite the issue of high computational complexity, these CNN-based approaches, proved to be effective and to have a high accuracy in detecting mitotic cells [12].…”

Section: Introductionmentioning

confidence: 99%

Automated mitosis detection in histopathology based on non-gaussian modeling of complex wavelet coefficients

et al. 2017

View full text Add to dashboard Cite

General rightsThis document is made available in accordance with publisher policies. Please cite only the published version using the reference above. AbstractTo diagnose breast cancer, the number of mitotic cells present in histology sections is an important indicator for examining and grading biopsy specimen. This study aims at improving the accuracy of automated mitosis detection by characterizing mitotic cells in wavelet based multi-resolution representations via a non-Gaussian modeling method. The potential mitosis candidates were decomposed into multi-scale forms by an undecimated dual-tree complex wavelet transform. Two non-Gaussian models (the generalized Gaussian distribution (GGD) and the symmetric alpha-stable (SαS) distributions) were used to accurately model the heavy-tailed behavior of wavelet marginal distributions. The method was evaluated on two independent data cohorts, including the benchmark dataset (MITOS), via a support vector machine classifier.

show abstract

Mitosis Detection in Breast Cancer Histology Images with Deep Neural Networks

Cited by 1,070 publications

References 16 publications

Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks

Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks

DGCC: data-driven granular cognitive computing

Automated mitosis detection in histopathology based on non-gaussian modeling of complex wavelet coefficients

Contact Info

Product

Resources

About