A survey on deep learning in medicine: Why, how and when?

doi:10.1016/j.inffus.2020.09.006

Information Fusion

Volume 66, February 2021, Pages 111-137

https://doi.org/10.1016/j.inffus.2020.09.006 Get rights and content

Highlights

•
We review the state-of-the-art focusing on the application of DL in medicine.
•
We expose a categorization of Deep Learning models used and applied in medicine.
•
We classify medicine-related DL applications into macro-areas and sub-areas.
•
We thoroughly discuss the recent and open challenges related to DL in medicine.

Abstract

New technologies are transforming medicine, and this revolution starts with data. Health data, clinical images, genome sequences, data on prescribed therapies and results obtained, data that each of us has helped to create. Although the first uses of artificial intelligence (AI) in medicine date back to the 1980s, it is only with the beginning of the new millennium that there has been an explosion of interest in this sector worldwide. We are therefore witnessing the exponential growth of health-related information with the result that traditional analysis techniques are not suitable for satisfactorily management of this vast amount of data. AI applications (especially Deep Learning), on the other hand, are naturally predisposed to cope with this explosion of data, as they always work better as the amount of training data increases, a phase necessary to build the optimal neural network for a given clinical problem. This paper proposes a comprehensive and in-depth study of Deep Learning methodologies and applications in medicine. An in-depth analysis of the literature is presented; how, where and why Deep Learning models are applied in medicine are discussed and reviewed. Finally, current challenges and future research directions are outlined and analysed.

Introduction

In the coming years, Artificial Intelligence (AI) will have an increasingly important role in the field of medicine, where it is already making a difference today. Medicine based on the observation of events has been, for many years, ever since the time of Hippocrates, the epistemological guiding criterion of the healthcare profession. This approach has evolved, with the progress of medicine, into what is termed Evidence-Based Medicine (EBM). Today, indeed, medicine based on signs which cannot be observed by any human doctor but can become evident with the use of Big Data and Deep Learning (DL) techniques has been developed. Such techniques are able to consider and process much more information than is possible for any human. State-of-the-art Deep Neural Networks (DNNs), also known as DL modes [1] have demonstrated remarkable results in image processing, classification and data analysis.

DL is increasingly attracting the interest of researchers in the medical and healthcare sectors, since, by using medical data, it is possible to increase the accuracy of medical applications. In particular, DL is rapidly replacing classic neural network techniques, named artificial neural networks (ANNs), whose goal is to mimic the human brain. This trend can be motivated by the following reasons. Firstly, DL can provide a better interpretation of a very complex phenomenon than classic statistical approaches if high-dimensional datasets are available, and the performance of a DL is directly proportional to the input size. This is a common scenario in medicine, where, as pointed out in [2], large amounts of data (about 15 to 20 TB) are collected and stored in optimized databases every day, also by using Cloud computing platforms [3]. Furthermore, DL is characterized by a high degree of flexibility. Medical data include different types of unstructured data, such as images, signals, genetic expressions and text data. Thanks to the complexity of their architectures, DL frameworks are able to benefit from this heterogeneity by achieving high levels of abstraction in data analysis. Finally, the high level of automation [4]. ML algorithms require a manual intervention to select the fundamental information from the input data and the corresponding transformation rules [5]. This is a crucial challenge because an experts’ decision is needed and, therefore, there is a corresponding increase in the time and costs for a diagnosis [6]. However, DL can determine these elements by using large samples of examples. There are two main consequences of this facility. Firstly, there is a significant reduction in the cost and time of treatment and diagnosis, Secondly, the independence of the diagnosis means that patients can talk directly to data scientists and, by running software, can understand the cause of their disease and obtain the best treatment.

The application of AI in real contexts can result in numerous potential advantages, such as the execution speed, potential reduction in costs, both direct and indirect, better diagnostic accuracy, greater clinical and operational efficiency (“algorithms don’t sleep”) and the possibility of providing access to the clinical information even for people who cannot otherwise benefit from this for geographical, political and economic reasons.

A great number of publications and surveys have addressed the use of DL in medicine, focusing on specific challenges or medical fields [7], [8], [9], [10]. Nevertheless, most of these works are lacking in details, difficult to compare and do not provide the reader with a comprehensive overview of the applications of DL in the general medical area. Fig. 1 presents a keywords-generated tree-map extracted from a Scopus¹ dataset composed of papers related to “Deep Learning” and “Medicine” as input words. From 2016 until now, more than 1200 papers have been considered within this dataset. The tree-map has been generated by using the bibliometrix R-package,² an open-source tool for quantitative research in scientometrics and bibliometrics that includes all the main bibliometric methods of analysis. By analysing the tree-map (starting from the left side and considering the biggest squares), it is evident that DL in Medicine is mostly applied in the task of image processing, with a great focus on diagnostics. By continuing the analysis towards the right, some crucial keywords can be observed, such as “aged”, “personalized medicine”, and “classification”. Summarizing the results of the keyword-based tree-map, it is possible to have an overview of the main medical fields, the principal tasks performed, and the most frequently used algorithms relating to DL in medicine during the last few years.

Starting from the above considerations, in this paper our aim is to provide an extensive analysis of DL applications in medicine, also categorizing DL models in relation to their applications in different medical areas. Afterwards, a comprehensive study of the state-of-the-art DL in medicine will be performed, taking into account existing DL surveys focused on specific medical fields. In summary, with this paper we aim to make the following contribution:

1.
We will review the state-of-the-art in papers and surveys, especially of recent years, focusing on the application of DL in medicine, including all medical areas.
2.
We will present a categorization of the DL models used and applied in medicine and give clear definitions of each, also providing an overview of hybrid DL architectures.
3.
We will comprehensively classify medicine-related DL applications into macro-areas, also describing their sub-areas and the key aspects of the applied DL models.
4.
We will analyse and discuss the recent and open challenges related to DL in medicine, also addressing future research directions, in order to provide the reader with a clear overview of the real-world scenario.

The rest of the paper is organized as follows, as depicted in Fig. 2. In Section 2, the various DL models are described and analysed in depth, with hybrid architectures also presented. In Section 3, a comprehensive overview and classification of DL applications in medicine is provided, including also a description of the main properties of the applied DL models. Section 4 presents a review of the kinds of medical data and hyperparameter optimization techniques. The current challenges in relation to the application of DL in medicine and future research directions are outlined in Section 5. Finally, in Section 6 our conclusions are presented.

Section snippets

Deep learning models

In this section we will provide a comprehensive overview of DL models applied in medicine. Starting from an in-depth study of the literature, we will present the main families of DL architectures: Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Autoencoders (AE), Generative Adversarial Networks (GAN), Deep Belief Networks (DBN) and Hybrid Architectures (HA) (see Fig. 3).

In the following subsections of this section, the various DL models will be discussed: Section 2.1 the

Applications fields of DL in medicine

DL achieves a remarkable accuracy and quality of its results thanks to its multi-layer architecture, which is able to obtain a high level of abstraction by working with large data samples. For this reason, it is gaining a great popularity in all fields where the process of information extraction from data involves various problems, such as the medical sector [93], [94], [95].

In detail, the analysis of medical data encounters three main issues, which are summarized in the following discussion.

Data structure and hyperparameters optimization

DL methodologies in medicine, as well as in all in the other fields where these techniques can be applied, often require the analysis of some “algorithmic” problems that can arise from data or from the algorithm itself. For these reasons, in the following Section a brief discussion on two of the main issues involved in the usage of Neural network approaches are discussed: on one side the format of stored data usually affect the class of algorithms that can be used on a specific problem, so in

Challenges and future research directions

DL is becoming the new paradigm in the analysis of medical data, as confirmed by the results discussed in Section 3. In addition, in recent years many other medical fields are beginning to benefit from the ability of these models to extract information from very different kinds of data. However, the complexity of DL models, the heterogeneity of medical data and the necessary interaction between machines and humans pose several issues, which must be taken into account in any assessment of future

Conclusions

DL is changing the cultural paradigm of medicine: its applications could become increasingly indispensable in terms of providing answers in contexts of high complexity and uncertainty and in order to allow doctors to have more time to take care of the medical needs of their patients. However, data are not values; any intervention based on data must be personalized, also taking into account the frequently contradictory nature of the knowledge provided in the literature. DL will be useful mainly

CRediT authorship contribution statement

Francesco Piccialli: Conceptualization, Methodology, Investigation, Writing, Visualization, Supervision. Vittorio Di Somma: Data curation, Writing, Investigation. Fabio Giampaolo: Writing, Resources, Review & editing. Salvatore Cuomo: Formal analysis, Writing - review & editing. Giancarlo Fortino: Writing - review & editing, Validation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors dedicate this work to their friend and colleague Prof. Antonio Picariello, who passed away prematurely. The memory of his wonderful and kind soul will always remain in our hearts.

This work was supported by the CUP-in-un-click (CUP-in-One-Click) research, Italy project [Regione Campania - Bando RIS3 2018 - Fase 2]. The authors would like to thank the M.O.D.A.L. research laboratory (http://www.labdma.unina.it/index.php/modal/) for their efforts and support.

References (305)

FortinoG. et al.
Bodycloud: A saas approach for community body sensor networks
Future Gener. Comput. Syst.
(2014)
GaoK. et al.
Julia language in machine learning: Algorithms, applications, and open issues
Comp. Sci. Rev.
(2020)
LitjensG. et al.
A survey on deep learning in medical image analysis
Med. Image Anal.
(2017)
MeyerP. et al.
Survey on deep learning for radiotherapy
Comput. Biol. Med.
(2018)
SchmidhuberJ.
Deep learning in neural networks: An overview
Neural Netw.
(2015)
HuZ. et al.
Deep learning for image-based cancer detection and diagnosis- a survey
Pattern Recognit.
(2018)
YuY. et al.
Deep transfer learning for modality classification of medical images
Information
(2017)
Al-antariM.A. et al.
Evaluation of deep learning detection and classification towards computer-aided diagnosis of breast lesions in digital x-ray mammograms
Comput. Methods Programs Biomed.
(2020)
ReddyB.K. et al.
Predicting hospital readmission for lupus patients: An RNN-LSTM-based deep-learning methodology
Comput. Biol. Med.
(2018)
PiccialliF. et al.
A deep learning approach for path prediction in a location-based IoT system
Pervasive Mob. Comput.
(2020)

PhamT. et al.

Predicting healthcare trajectories from medical records: A deep learning approach

J. Biomed. Inform.

(2017)

AckleyD.H. et al.

A learning algorithm for Boltzmann machines

Cogn. Sci.

(1985)

RibeiroM. et al.

A study of deep convolutional auto-encoders for anomaly detection in videos

Pattern Recognit. Lett.

(2018)

HintonG.E.

Learning multiple layers of representation

Trends Cogn. Sci.

(2007)

HassanM.M. et al.

Human emotion recognition using deep belief network architecture

Inf. Fusion

(2019)

ZreikM. et al.

Deep learning analysis of the myocardium in coronary CT angiography for identification of patients with functionally significant coronary artery stenosis

Med. Image Anal.

(2018)

CaballoM. et al.

Deep learning-based segmentation of breast masses in dedicated breast CT imaging: Radiomic feature stability between radiologists and artificial intelligence

Comput. Biol. Med.

(2020)

ZhaoR. et al.

Deep learning and its applications to machine health monitoring

Mech. Syst. Signal Process.

(2019)

KulkarniS. et al.

Artificial intelligence in medicine: where are we now?

Acad. Radiol.

(2020)

GoodfellowI. et al.

Deep Learning

(2016)

YueL. et al.

Deep learning for heterogeneous medical data analysis

World Wide Web

(2020)

EstevaA. et al.

A guide to deep learning in healthcare

Nat. Med.

(2019)

ChenD. et al.

Deep learning and alternative learning strategies for retrospective real-world clinical data

NPJ Digit. Med.

(2019)

AkayA. et al.

Deep learning: current and emerging applications in medicine and technology

IEEE J. Biomed. Health Inf.

(2019)

ZhangZ. et al.

Deep learning in omics: a survey and guideline

Brief. Funct. Genom.

(2019)

NielsenM.A.

Neural Networks and Deep Learning, Vol. 2018

(2015)

LeCunY. et al.

Handwritten digit recognition with a back-propagation network

ChengG. et al.

When deep learning meets metric learning: Remote sensing image scene classification via learning discriminative CNNs

IEEE Trans. Geosci. Remote Sens.

(2018)

ZhaoZ.-Q. et al.

Object detection with deep learning: A review

IEEE Trans. Neural Netw. Learn. Syst.

(2019)

ChenL.-C. et al.

Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs

IEEE Trans. Pattern Anal. Mach. Intell.

(2017)

RawatW. et al.

Deep convolutional neural networks for image classification: A comprehensive review

Neural Comput.

(2017)

AhmedI. et al.

Exploring deep learning models for overhead view multiple object detection

IEEE Internet Things J.

(2020)

KarpathyA. et al.

Large-scale video classification with convolutional neural networks

KrizhevskyA. et al.

Imagenet classification with deep convolutional neural networks

RonnebergerO. et al.

U-net: Convolutional networks for biomedical image segmentation

AkhtarN. et al.

Interpretation of intelligence in CNN-pooling processes: a methodological survey

Neural Comput. Appl.

(2019)

J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in: Proceedings of the IEEE...

N. Suda, V. Chandra, G. Dasika, A. Mohanty, Y. Ma, S. Vrudhula, J.-s. Seo, Y. Cao, Throughput-optimized OpenCL-based...

SimonyanK. et al.

Very deep convolutional networks for large-scale image recognition

MengD. et al.

Liver fibrosis classification based on transfer learning and FCNet for ultrasound images

Ieee Access

(2017)

TangY. et al.

Scene text detection and segmentation based on cascaded convolution neural networks

IEEE Trans. Image Process.

(2017)

K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: Proceedings of the IEEE Conference...

F. Wang, M. Jiang, C. Qian, S. Yang, C. Li, H. Zhang, X. Wang, X. Tang, Residual attention network for image...

HersheyS. et al.

CNN architectures for large-scale audio classification

DongH. et al.

Automatic brain tumor detection and segmentation using u-net based fully convolutional networks

ÇiçekÖ. et al.

3D U-Net: learning dense volumetric segmentation from sparse annotation

ZhouZ. et al.

Unet++: A nested u-net architecture for medical image segmentation

BadrinarayananV. et al.

Segnet: A deep convolutional encoder-decoder architecture for image segmentation

IEEE Trans. Pattern Anal. Mach. Intell.

(2017)

GirshickR. et al.

Rich feature hierarchies for accurate object detection and semantic segmentation

DaiJ. et al.

R-fcn: Object detection via region-based fully convolutional networks

Cited by (226)

Multiscale triplet spatial information fusion-based deep learning method to detect retinal pigment signs with fundus images
2024, Engineering Applications of Artificial Intelligence
Inherited retinal diseases (IRDs) are genetic disorders that cause progressive deterioration of the photoreceptors associated with vision loss or blindness. Retinitis pigmentosa (RP) is a rare hereditary ophthalmic disease that initially causes night blindness owing to continuous retinal pigment deterioration. A computer-aided diagnosis (CAD)-based RP diagnosis solution by pigment sign detection can help ophthalmologists to analyze and treat the disease timely. At present, most of the research addresses retinal disease CAD using expensive optical coherence tomography (OCT); however, fundus imaging-based solutions are quick, convenient, and inexpensive for massive screening. This study proposes two convolutional neural networks (CNNs)-based segmentation that combines multiscale features by spatial information fusion: a single spatial fusion network (SSF-Net) and a triplet spatial fusion network (TSF-Net). SSF-Net fuses four multiscale spatial information streams. TSF-Net exploits triplet spatial information fusion by early, intermediate, and late fusion to ensure the fine segmentation of retinal pigment signs without preprocessing. TSF-Net creates a valuable difference in performance over SSF-Net. To evaluate SSF-Net and TSF-Net, the open dataset, named Retinal Images for Pigment Signs is utilized with 4-fold cross-validation. The experiment results confirm that SSF-Net and TSF-Net demonstrate superior performance compared to the state-of-the-art methods for the screening and analysis of RP disease.
Contour-induced parallel graph reasoning for liver tumor segmentation
2024, Biomedical Signal Processing and Control
The accurate detection and segmentation of liver cancers from abdominal CT scans is critical. However, segmenting liver tumors presents significant hurdles due to indistinct lesion boundaries and ignoring the correlation between target and outlines.
In this paper, we propose the Parallel Graph Convolutional Network (PGC-Net), a completely novel segmentation framework for liver tumors. With regard to segmentation against constraints, we specifically use contour-induced parallel graph reasoning for quick yet efficient segmentation. First, we use a Pyramid Vision Transformer that has already been trained to extract multi-scale features of region and contour. In order to project the pixels into two distinct high-dimensional areas, we secondly use the parallel graph reasoning strategy, where the vertices are weighted in accordance with the geometric prior of the contour. Through the process of graph convolution, the complementary properties of region and contour also propagate the information. Finally, we project back to the original pixel space for the prediction using the refined features deduced from the graph.
Experimental results on two available datasets, LiTS17 (with an average Dice score of 73.63%) and 3DIRCADb (with an average Dice score of 74.16%).
Our framework focused on the interaction between two orthogonal graphs and contour information, which has the potential to improve the accuracy and efficiency of liver tumor segmentation.
Fusion of standard and ordinal dropout techniques to regularise deep models
2024, Information Fusion
Dropout is a popular regularisation tool for deep neural classifiers, but it is applied regardless of the nature of the classification task: nominal or ordinal. Consequently, the order relation between the class labels of ordinal problems is ignored. In this paper, we propose the fusion of standard dropout and a new dropout methodology for ordinal classification regularising deep neural networks to avoid overfitting and improve generalisation, but taking into account the extra information of the ordinal task, which is exploited to improve performance. The correlation between the outputs of every neuron and the target labels is used to guide the dropout process: the higher the neuron is correlated with the expected labels, the lower its probability of being dropped. Given that randomness also plays a crucial role in the regularisation process, a balancing factor ( $β$ ) is also added to the training process to determine the influence of the ordinality with respect to a constant probability, providing a hybrid ordinal regularisation method. An extensive battery of experiments shows that the new hybrid ordinal dropout methodology perform better than standard dropout, obtaining improved results in most evaluation metrics, including not only ordinal metrics but also nominal ones.
On the search for efficient face recognition algorithm subject to multiple environmental constraints
2024, Heliyon
From literature, majority of face recognition modules suffer performance challenges when presented with test images acquired under multiple constrained environments (occlusion and varying expressions). The performance of these models further deteriorates as the degree of degradation of the test images increases (relatively higher occlusion level). Deep learning-based face recognition models have attracted much attention in the research community as they are purported to outperform the classical PCA-based methods. Unfortunately their application to real-life problems is limited because of their intensive computational complexity and relatively longer run-times. This study proposes an enhancement of some PCA-based methods (with relatively lower computational complexity and run-time) to overcome the challenges posed to the recognition module in the presence of multiple constraints. The study compared the performance of enhanced classical PCA-based method (HE-GC-DWT-PCA/SVD) to FaceNet algorithm (deep learning method) using expression variant face images artificially occluded at 30% and 40%. The study leveraged on two statistical imputation methods of MissForest and Multiple Imputation by Chained Equations (MICE) for occlusion recovery. From the numerical evaluation results, although the two models achieved the same recognition rate (85.19%) at 30% level of occlusion, the enhanced PCA-based algorithm (HE-GC-DWT-PCA/SVD) outperformed the FaceNet model at 40% occlusion rate, with a recognition rate of 83.33%. Although both Missforest and MICE performed creditably well as de-occlusion mechanisms at higher levels of occlusion, MissForest outperforms the MICE imputation mechanism. MissForest imputation mechanism and the proposed HE-GC-DWT-PCA/SVD algorithm are recommended for occlusion recovery and recognition of multiple constrained test images respectively.
Sparse Dynamic Volume TransUNet with multi-level edge fusion for brain tumor segmentation
2024, Computers in Biology and Medicine
3D MRI Brain Tumor Segmentation is of great significance in clinical diagnosis and treatment. Accurate segmentation results are critical for localization and spatial distribution of brain tumors using 3D MRI. However, most existing methods mainly focus on extracting global semantic features from the spatial and depth dimensions of a 3D volume, while ignoring voxel information, inter-layer connections, and detailed features. A 3D brain tumor segmentation network SDV-TUNet (Sparse Dynamic Volume TransUNet) based on an encoder–decoder architecture is proposed to achieve accurate segmentation by effectively combining voxel information, inter-layer feature connections, and intra-axis information. Volumetric data is fed into a 3D network consisting of extended depth modeling for dense prediction by using two modules: sparse dynamic (SD) encoder–decoder module and multi-level edge feature fusion (MEFF) module. The SD encoder–decoder module is utilized to extract global spatial semantic features for brain tumor segmentation, which employs multi-head self-attention and sparse dynamic adaptive fusion in a 3D extended shifted window strategy. In the encoding stage, dynamic perception of regional connections and multi-axis information interactions are realized through local tight correlations and long-range sparse correlations. The MEFF module achieves the fusion of multi-level local edge information in a layer-by-layer incremental manner and connects the fusion to the decoder module through skip connections to enhance the propagation ability of spatial edge information. The proposed method is applied to the BraTS2020 and BraTS2021 benchmarks, and the experimental results show its superior performance compared with state-of-the-art brain tumor segmentation methods. The source codes of the proposed method are available at https://github.com/SunMengw/SDV-TUNet.
FaceNet recognition algorithm subject to multiple constraints: Assessment of the performance
2024, Scientific African
Literature has it that the performance of most face recognition algorithms still decline in multiple constrained environments (Occlusions and Expressions), despite the achieved successes of deep learning face recognition algorithms. Using expression variant test face images synthetically occluded at 30% and 40% rates, the study evaluated the performance of FaceNet deep learning model for face recognition under the aforementioned constraints and when three (3) statistical multiple imputation methods (Multivariable Imputation using Chain Equations (MICE), MissForest and Regularized Expectation Maximization (RegEM)) are adopted for occlusion recovery. Results of the study showed improved recognition rates of the study algorithm when the imputation-based recovered faces were used for recognition compared with using their multiple constrained counterparts. However, test faces reconstructed with the MissForest imputation method were more accurately recognized using the FaceNet deep learning algorithm. Furthermore, the study demonstrated that some simple augmentation schemes sufficed to further enhance the performance of the FaceNet model. Specifically, the FaceNet algorithms gave the highest average recognition rates (85.19% and 79.5% for 30% and 40% occlusion levels respectively) under augmentation scheme IV (slight rotations, horizontal flipping, shearing, brightness adjustments, and stretching) using MissForest as the de-occlusion mechanism. The study also found that, no disparity existed in its performance with the choice of either Support Vector Machines (SVM) or City Block (CB) for classification under augmentation scheme IV. The study recommends using the MissForest imputation method in dealing with moderately high occluded test faces with varying expressions to enhance the performance of the FaceNet face recognition model.

View all citing articles on Scopus

View full text

Full length articleA survey on deep learning in medicine: Why, how and when?

Highlights

Abstract

Introduction

Section snippets

Deep learning models

Applications fields of DL in medicine

Data structure and hyperparameters optimization

Challenges and future research directions

Conclusions

CRediT authorship contribution statement

Declaration of Competing Interest

Acknowledgements

Future Gener. Comput. Syst.

Comp. Sci. Rev.

Med. Image Anal.

Comput. Biol. Med.

Neural Netw.

Pattern Recognit.

Information

Comput. Methods Programs Biomed.

Comput. Biol. Med.

Pervasive Mob. Comput.

J. Biomed. Inform.

Cogn. Sci.

Pattern Recognit. Lett.

Trends Cogn. Sci.

Inf. Fusion

Med. Image Anal.

Comput. Biol. Med.

Mech. Syst. Signal Process.

Acad. Radiol.

Deep Learning

Deep learning for heterogeneous medical data analysis

World Wide Web

A guide to deep learning in healthcare

Nat. Med.

Deep learning and alternative learning strategies for retrospective real-world clinical data

NPJ Digit. Med.

Deep learning: current and emerging applications in medicine and technology

IEEE J. Biomed. Health Inf.

Deep learning in omics: a survey and guideline

Brief. Funct. Genom.

Neural Networks and Deep Learning, Vol. 2018

Handwritten digit recognition with a back-propagation network

When deep learning meets metric learning: Remote sensing image scene classification via learning discriminative CNNs

IEEE Trans. Geosci. Remote Sens.

Object detection with deep learning: A review

IEEE Trans. Neural Netw. Learn. Syst.

Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs

IEEE Trans. Pattern Anal. Mach. Intell.

Deep convolutional neural networks for image classification: A comprehensive review

Neural Comput.

Exploring deep learning models for overhead view multiple object detection

IEEE Internet Things J.

Large-scale video classification with convolutional neural networks

Imagenet classification with deep convolutional neural networks

U-net: Convolutional networks for biomedical image segmentation

Interpretation of intelligence in CNN-pooling processes: a methodological survey

Neural Comput. Appl.

Very deep convolutional networks for large-scale image recognition

Liver fibrosis classification based on transfer learning and FCNet for ultrasound images

Ieee Access

Scene text detection and segmentation based on cascaded convolution neural networks

IEEE Trans. Image Process.

CNN architectures for large-scale audio classification

Automatic brain tumor detection and segmentation using u-net based fully convolutional networks

3D U-Net: learning dense volumetric segmentation from sparse annotation

Unet++: A nested u-net architecture for medical image segmentation

Segnet: A deep convolutional encoder-decoder architecture for image segmentation

IEEE Trans. Pattern Anal. Mach. Intell.

Rich feature hierarchies for accurate object detection and semantic segmentation

R-fcn: Object detection via region-based fully convolutional networks

Full length article
A survey on deep learning in medicine: Why, how and when?