Multi-resolution-Tract CNN with Hybrid Pretrained and Skin-Lesion Trained Layers

Kawahara, Jeremy; Hamarneh, Ghassan

doi:10.1007/978-3-319-47157-0_20

Cited by 128 publications

(76 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A straightforward extension to classic feature extraction is to use deep learning for feature extraction combined with conventional machine learning methods for skin lesion classification [11], [12]. More recent approaches moved to endto-end trainable convolutional neural networks (CNNs) for lesion diagnosis [13]- [15]. In addition, multi-modal approaches using clinical images, dermoscopy and meta data have been proposed [16], as well as a method where segmentation and lesion structure information is incorporated into the system [17].…”

Section: Introductionmentioning

confidence: 99%

“…However, for skin lesion diagnosis, the relevant ROI is not necessarily known as image-level labels are used instead of pixel-level labels. Kawahara et al [13] used a two-path CNN with two input resolutions of the entire lesion image. Due to the high image resolution, this requires extensive downsampling or pooling inside the CNN's highresolution path which, again, might remove relevant features.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Skin Lesion Classification Using CNNs With Patch-Based Attention and Diagnosis-Guided Loss Weighting

Gessert

Sentker

Madesta

et al. 2020

IEEE Trans. Biomed. Eng.

123

View full text Add to dashboard Cite

Objective: This work addresses two key problems of skin lesion classification. The first problem is the effective use of high-resolution images with pretrained standard architectures for image classification. The second problem is the high class imbalance encountered in real-world multi-class datasets. Methods: To use high-resolution images, we propose a novel patch-based attention architecture that provides global context between small, high-resolution patches. We modify three pretrained architectures and study the performance of patch-based attention. To counter class imbalance problems, we compare oversampling, balanced batch sampling, and class-specific loss weighting. Additionally, we propose a novel diagnosis-guided loss weighting method which takes the method used for groundtruth annotation into account. Results: Our patch-based attention mechanism outperforms previous methods and improves the mean sensitivity by 7 %. Class balancing significantly improves the mean sensitivity and we show that our diagnosis-guided loss weighting method improves the mean sensitivity by 3 % over normal loss balancing. Conclusion: The novel patch-based attention mechanism can be integrated into pretrained architectures and provides global context between local patches while outperforming other patch-based methods. Hence, pretrained architectures can be readily used with high-resolution images without downsampling. The new diagnosis-guided loss weighting method outperforms other methods and allows for effective training when facing class imbalance. Significance: The proposed methods improve automatic skin lesion classification. They can be extended to other clinical applications where high-resolution image data and class imbalance are relevant.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Skin Lesion Classification Using CNNs With Patch-Based Attention and Diagnosis-Guided Loss Weighting

Gessert

Sentker

Madesta

et al. 2020

IEEE Trans. Biomed. Eng.

123

View full text Add to dashboard Cite

show abstract

“…In the literature, we have seen that many researchers have used multimodel CNN architectures to tackle this problem. For skin lesion classification, Kawahara and Hamarneh [16] presented a novel CNN architecture that is composed of multiple tracts, which is each tract analyzes the image at a different resolution simultaneously. For skin lesion classification designed to learn based on information from multiple image resolutions while leveraging pre-trained CNNs.…”

Section: Medical Image Processingmentioning

confidence: 99%

“…Although DL approaches have been widely used in many Medical Image Processing and achieved successful results [8,13,14,16,20,66,170,171,176,180,183,223], it seems that they have not been used much in vessel segmentation yet. In literature, we observed that there are only a few studies to determine the eye vessels [224][225][226].…”

Section: Suggestions On Deep Learning (Derin öğRenme öNerileri)mentioning

confidence: 99%

Untitled

2019

GMBD

View full text Add to dashboard Cite

Derin öğrenme (Deep Learning-DL), birçok alanda önemli başarılar elde etmiş güçlü bir makine öğrenmesi yöntemidir. Özellikle son on yılda, bilgisayarlı görü, nesne tanıma, konuşma tanıma, doğal dil işleme gibi birçok araştırma alanında başarılı sonuçlar elde ederek, yapay zekanın derin uykudan uyanmasına yol açmıştır. Günümüzde, çeşitli alanlardaki birçok araştırmacı, DL yöntemlerini kullanarak alanlarında en iyi sonucu almaya çalışmaktadır. Bu tarama çalışmasında, DL modelleri ve DL ile çalışılabilecek önemli araştırma konuları hakkında bilgiler vererek araştırmacılara rehberlik etmeyi hedefliyoruz. Çalışmada Özerk Araçlar (Autonomous Vehicles), Doğal Dil İşleme (Natural Language Processing), El Yazısı Karakter Tanıma (Handwritten Character Recognition), İmza Doğrulama (Signature Verification), Ses ve Video Tanıma (Voice and Video Recognition), Tıbbi Görüntü İşleme (Medical İmage Processing), Büyük Veri (Big Data) gibi dünyanın en popüler ve en zorlu alanlarında yapılan DL çalışmalarını inceliyoruz. Ayrıca, araştırmacılara yardımcı olmak için, incelediğimiz bu alanlardaki DL ile çalışılabilecek, henüz çalışılmamış veya yeterince iyi sonuçlar elde edilememiş problemlere dikkat çekerek olası araştırma konularını listeliyoruz. Günümüzde, Derin öğrenme yöntemleri nesne tanıma alanında insanlardan daha iyi sonuçlara ulaşmıştır. DL üzerine yapılan çalışmalar dikkate alındığında, bu başarının otonom araçlar, tıbbi görüntü işleme, büyük veri analizi ve karakter tanıma gibi alanlarda da gerçekleşeceği öngörülmektedir. https://dx.Deep learning (DL) is an important machine learning field that has achieved considerable success in many research areas. In the last decade, the-state-of-the-art studies on many research areas such as computer vision, object recognition, speech recognition, and natural language processing were especially led to the awakening of the artificial intelligence from deep sleep. Nowadays, many researchers try to find solutions to many problems in various fields under the light of DL methods. In this study, it is presented important knowledge to guide about DL models and challenging topics that can be used in DL for researchers. This study investigated DL studies which are made in the most popular and challenging fields such as autonomous vehicles, natural language processing, handwritten character recognition, signature verification, voice and video recognition, medical image processing, and big data. Furthermore, this study points out the remaining challenges of these research areas that can be solved by DL, and discusses future topics to help the researchers. In the present day, Deep learning methods have reached better results than humans in object recognition. According to the literature studies on DL, It is foreseen that this success will be achieved in areas such as autonomous vehicles, medical image processing, big data analysis, and character recognition. https://dx.

show abstract

“…To help physicians achieve high diagnostic accuracy, many assistant systems were proposed. Many diseases, including glaucoma [2], skin cancer [3], breast cancer [4], and leukemia [5], are already addressed by such systems. Early and accurate diagnoses could effectively reduce treatment costs, increase the probability of remission, or even prolong the lives of patients [1].…”

Section: Introductionmentioning

confidence: 99%

Deep Transfer Learning in Diagnosing Leukemia in Blood Cells

2020

View full text Add to dashboard Cite

Leukemia is a fatal disease that threatens the lives of many patients. Early detection can effectively improve its rate of remission. This paper proposes two automated classification models based on blood microscopic images to detect leukemia by employing transfer learning, rather than traditional approaches that have several disadvantages. In the first model, blood microscopic images are pre-processed; then, features are extracted by a pre-trained deep convolutional neural network named AlexNet, which makes classifications according to numerous well-known classifiers. In the second model, after pre-processing the images, AlexNet is fine-tuned for both feature extraction and classification. Experiments were conducted on a dataset consisting of 2820 images confirming that the second model performs better than the first because of 100% classification accuracy.Computers 2020, 9, 29 2 of 12 and classification. However, the success of each step depends on the success of the preceding step. For example, the success of classification depends on the success of the preceding feature extraction, which itself depends on the success of the preceding segmentation. Hence, high classification accuracy requires the success of all steps, each of which is non-trivial and problem-dependent [1,7].Recently, deep learning achieved many breakthroughs in different fields such as computer vision, natural language processing, and object recognition. Deep neural networks, such as convolutional neural networks (CNNs), can be used effectively to build computer-aided diagnostic systems. However, the design and training of deep neural networks are non-trivial and time-consuming tasks. Hence, instead of building a deep neural network from scratch, we use the concept of transfer learning, in which a deep network that achieved success in solving a certain problem is tuned to solve another problem.This paper proposes two classification models that are based on transfer learning and can distinguish between healthy and unhealthy blood smear images with high accuracy. These models employ AlexNet, which is a deep CNN that achieved huge success in the image classification challenge, ImageNet 2012.The remaining sections of this paper are arranged as follows: Section 2 covers some of the traditional and end-to-end-based methods described in the literature. Section 3 presents our two proposed classification models. Section 4 discusses their implementation in our experiments. Section 5 discusses the results obtained. Finally, Section 6 concludes the paper.

show abstract

Multi-resolution-Tract CNN with Hybrid Pretrained and Skin-Lesion Trained Layers

Cited by 128 publications

References 7 publications

Skin Lesion Classification Using CNNs With Patch-Based Attention and Diagnosis-Guided Loss Weighting

Skin Lesion Classification Using CNNs With Patch-Based Attention and Diagnosis-Guided Loss Weighting

Untitled

Deep Transfer Learning in Diagnosing Leukemia in Blood Cells

Contact Info

Product

Resources

About