Flexible architectures for retinal blood vessel segmentation in high-resolution fundus images

Bendaoudi, Hamza; Cheriet, Foued; Manraj, Ashley; Tahar, Houssem Ben; Langlois, J. M. Pierre

doi:10.1007/s11554-016-0661-4

Cited by 8 publications

(8 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore unsupervised methods are still dominant in works that focus on embedded or mobile systems and on execution speed [6,8,9,24]. These methods often rely on matched filtering, contour tracing and morphological transformation techniques.…”

Section: Related Workmentioning

confidence: 99%

“…In [6], a combination of matched filtering and contour tracing is proposed and implemented on GPU, realizing an execution time of 10ms on DRIVE. FPGA implementations of matched filtering based methods are introduced in [8] and [24] that accomplish execution times of 2ms and 52.3 ms respectively. In addition to morphological transformations, Bibiloni et al [9] use CLAHE and hysteresis thresholding to segment the retinal vasculature and report single-core execution speeds of 37ms on DRIVE, using a Intel i5-3340 CPU.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

M2U-Net: Effective and Efficient Retinal Vessel Segmentation for Real-World Applications

Laibacher

Weyde

Jalali

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

In this paper, we present a novel neural network architecture for retinal vessel segmentation that improves over the state of the art on two benchmark datasets, is the first to run in real time on high resolution images, and its small memory and processing requirements make it deployable in mobile and embedded systems.The M2U-Net has a new encoder-decoder architecture that is inspired by the U-Net. It adds pretrained components of MobileNetV2 in the encoder part and novel contractive bottleneck blocks in the decoder part that, combined with bilinear upsampling, drastically reduce the parameter count to 0.55M compared to 31.03M in the original U-Net.We have evaluated its performance against a wide body of previously published results on three public datasets. On two of them, the M2U-Net achieves new state-of-the-art performance by a considerable margin. When implemented on a GPU, our method is the first to achieve real-time inference speeds on high-resolution fundus images. We also implemented our proposed network on an ARM-based embedded system where it segments images in between 0.6 and 15 sec, depending on the resolution. Thus, the M2U-Net enables a number of applications of retinal vessel structure extraction, such as early diagnosis of eye diseases, retinal biometric authentication systems, and robot assisted microsurgery.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

M2U-Net: Effective and Efficient Retinal Vessel Segmentation for Real-World Applications

Laibacher

Weyde

Jalali

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

show abstract

“…Early retinal processing work by Köhler and colleagues 117 used the retinal vessel contrast as a proxy measure for image quality, which was implemented later as fast real-time algorithm by Bendaoudi and colleagues. 118 Saha and colleagues 119 developed a structure-agnostic data-driven deep learning network for flagging fundus images either as acceptable for diabetic retinopathy screening or as to be recaptured. In practice, however, the cost function used for deep learning training can be defined in multiple ways as reviewed by Zhao and colleagues.…”

Section: Embedded Ophthalmic Devicesmentioning

confidence: 99%

Embedded deep learning in ophthalmology: making ophthalmic imaging smarter

et al. 2019

View full text Add to dashboard Cite

Deep learning has recently gained high interest in ophthalmology due to its ability to detect clinically significant features for diagnosis and prognosis. Despite these significant advances, little is known about the ability of various deep learning systems to be embedded within ophthalmic imaging devices, allowing automated image acquisition. In this work, we will review the existing and future directions for ‘active acquisition’–embedded deep learning, leading to as high-quality images with little intervention by the human operator. In clinical practice, the improved image quality should translate into more robust deep learning–based clinical diagnostics. Embedded deep learning will be enabled by the constantly improving hardware performance with low cost. We will briefly review possible computation methods in larger clinical systems. Briefly, they can be included in a three-layer framework composed of edge, fog, and cloud layers, the former being performed at a device level. Improved egde-layer performance via ‘active acquisition’ serves as an automatic data curation operator translating to better quality data in electronic health records, as well as on the cloud layer, for improved deep learning–based clinical data mining.

show abstract

“…Last but not least, advanced driver assistance systems in the automotive domain, and recent achievements for autonomous driving would not be possible without real-time image processing. This special issue collects recent research results within this domain of research and presents scientific achievements regarding the algorithms and architectures for the design of real-time image-processing systems.In this special issue, contributions from the domain of applications such as motion estimation hardware for H.264 multi-view video coding [1], a hardware implementation for traffic road sign detection [2] and identification system and retinal blood vessel segmentation in high-resolution fundus images [3] are collected. An additional contribution in this domain can be found in the manuscript [4], * Michael Huebner…”

mentioning

confidence: 99%

“…In this special issue, contributions from the domain of applications such as motion estimation hardware for H.264 multi-view video coding [1], a hardware implementation for traffic road sign detection [2] and identification system and retinal blood vessel segmentation in high-resolution fundus images [3] are collected. An additional contribution in this domain can be found in the manuscript [4], * Michael Huebner…”

mentioning

confidence: 99%

Special issue: Design and architectures for real-time signal and image processing

Huebner

2018

J Real-Time Image Proc

View full text Add to dashboard Cite

, image and control processing based on a new approach called Algorithm-Architecture-Matching, which aims to leverage the design flow by a simultaneous study of both algorithmic and architectural issues, taking into account multiple design constraints, as well as algorithm and architecture optimizations.The goal of DASIP is to present the latest results in the domain of design and architecture for signal and image processing along several axes: methods and tools; development platforms, architectures and technologies; use-cases and applications as well as smart-sensing systems.The best papers from the conference were invited to submit considerable extended manuscripts to this special issue with a topic in relation to the paper they presented during the conference. All the papers which appear in this special issue went through a peer review process.Embedded Systems and Cyber-physical Systems (CPS), especially in relation to image and signal processing, are a big step forward from the basic data-gathering networks, and they have been called to be the fourth industrial revolution. Therefore, the important contribution of CPS is its integration and connection of the hardware elements, e.g., sensors, actuators and computing units, with the software functionalities, e.g., automated routines and supervisor software, to perform energy efficiency and simultaneously keeping safety requirements. Many fields employ CPS into their operations, e.g., transportation, defense, energy and industrial automation, health and biomedical, agriculture and critical infrastructure. To automate the processes, the data gathering and the actions they can perform, while also lowering the quantity of human-made errors during run time. Furthermore, critical applications that must deal with several challenges find in CPS a powerful ally, since the system's level of automation and independence is higher than those of basic manually supervised networks.CPS supports directly the agenda of Industry 4.0, which is a key driver of Europe's economy and growth in industrial production, industrial automation and the related logistics. Only with novel concepts for sensors in a network, which goes far beyond what the current state of the art can deliver, the current and future requirements in terms of flexibility, safety and security can be provided. The leading industries in this area need novel solutions to develop energy efficient, highly flexible, robust, safe and secure sensor nodes for the next generation of factories in the age of Industry 4.0 with a holistic toolset. The key elements here are to increase efficiency, decrease time to market and to enhance flexibility. The Industrial Internet offers industry players the potential for high growth and improved efficiency.Real-time image processing plays a major role in several applications in the academic and industrial field. As an example, the application in a modern industrial environment includes camera data to handle the production of complex components. In robotic applications, real-time image pro...

show abstract

Flexible architectures for retinal blood vessel segmentation in high-resolution fundus images

Cited by 8 publications

References 29 publications

M2U-Net: Effective and Efficient Retinal Vessel Segmentation for Real-World Applications

M2U-Net: Effective and Efficient Retinal Vessel Segmentation for Real-World Applications

Embedded deep learning in ophthalmology: making ophthalmic imaging smarter

Special issue: Design and architectures for real-time signal and image processing

Contact Info

Product

Resources

About