Le 13/02/2021 par lmoalic :
Keywords: Artificial intelligence, deep learning, medical imaging, optimization algorithms, hybrids methods.
Context:
Machine learning has witnessed a tremendous amount of attention over the last few years. Deep neural networks are now the state-of-the-art machine learning models across a variety of areas, from image analysis to natural language processing, and widely deployed in academia and industry. These developments have a huge potential for medical imaging technology, medical data analysis, medical diagnostics and healthcare in general, slowly being realized. Accordingly, the deep learning algorithm gets a lot of attention these days to solve various problems in medical imaging fields [1]. One example is to detect disease or abnormalities from X-ray images and classify them into several disease types or severities in radiology. More interestingly, there has been an increasing interest in the role of imaging for diagnosis of COVID-19. This kind of task has been executed based on the various machine learning models, theoretical or empirical approaches. However, the proposed models might generate more false positives than physicians and thus lead to the increment of assessment time and unnecessary biopsies [2]. This is due to the fact that proposed models in the literature are designed to find solutions for some specific tasks and it is not possible to apply an existing machine learning pipeline to the medical domain and still have superior results in a simple straightforward manner [3]. For example, medical images contain various sizes and types of complex disease patterns and it would be difficult for standard models to be directly trained on such complicated disease patterns. Moreover, IBM researchers estimate that medical images currently account for at least 90 percent of all medical data, making it the largest data source in the healthcare industry. This becomes an overwhelming amount on a human scale, when you consider that radiologists in some hospital emergency rooms are presented with thousands of images daily. New methods are thus required to extract and represent data from those images more efficiently. Thus, a better classification strategy is needed for these large datasets.
Objectives:
The above-mentioned problem calls the need for a research study so as to find more accurate classifiers that can effectively be used to explore the medical imaging domain. Hence, we have to construct a specialized machine learning pipeline for medical imaging problems. The extra degree of freedom from the design space could make this process very time-consuming and demand for automated machine learning methods that can be adopted easily without any expert knowledge. Thanks to the meta-heuristics, these problems could be overcome with great answering accuracy and allow humans to spend time on other productive tasks. In this direction, we propose to find a highly efficient deep neural network that is optimized using meta-heuristic algorithms for medical imaging [4,5,6]. The deep networks use a hierarchy of features in conjunction with several layers to learn complex non-linear mappings between the input and output layers. At the opposite of traditional machine learning methods that use handmade features, the important features are discovered automatically and are represented hierarchically. This is known to be the strong point of deep networks against traditional machine learning approaches. Accordingly, these models have been described as universal learning approaches that are not task-specific and can be used to tackle different problems arising in different research domains. Particularly, we are interested in convolutional neural networks; which are regularized versions of fully-connected neural networks inspired from biological visual systems.
References:
Lundervold, Alexander Selvikvåg, and Arvid Lundervold. "An overview of deep learning in medical imaging focusing on MRI." Zeitschrift für Medizinische Physik 29.2 (2019): 102-127.
Lehman CD, Wellman RD, Buist DS, et al. Diagnostic accuracy of digital screening mammography with and without computer-aided detection. JAMA Intern Med. 2015;175:1828?37.
M.-A. Zöller & M. F. Huber. Survey on automated machine learning. CoRR, abs/1904.12054, 2019.
H. Rakhshani, H. Ismail-Fawaz, L. Idoumghar, G. Forestier, J. Lepagnot, J. Weber, M. Brevilliers and P.A Muller, ?Neural Architecture Search for Time Series Classification?. The International Joint Conference on Neural Networks (IJCNN), 19 - 24th July, 2020, Glasgow (UK) (http://www.mage.fst.uha.fr/idoumghar/ijcnn2020.pdf).
H. Rakhshani, B. Latard, M. Brevilliers, J. Weber, J. Lepagnot, G. Forestier, M. Hassenforder and L. Idoumghar, ?Automated Machine Learning for Information Retrieval in Scientific Articles?. IEEE Congress on Evolutionary Computation, 19 - 24th July, 2020, Glasgow (UK) (www.mage.fst.uha.fr/idoumghar/cec2020.pdf)
H. Rakhshani, L. Idoumghar, J. Lepagnot, M. Brévilliers, Ed. Keedwell ?Automatic hyperparameter selection in Autodock?. IEEE International Conference on Bioinformatics and Biomedicine, pp. 34-738, 3-6 Dec. 2018, Madrid, Spain.
Your Profile:
Master of Science in Computer Science or any discipline relevant to this area of research.
The candidate should have some knowledge and experience in the optimization, computer vision, and/or swarm intelligence domains.
Strong programming skills in Matlab, Python, Java, or C++.
The candidate should be fluent in English and/or French languages.
Contacts and Application:
Applicants are invited to contact the advisors: lhassane.idoumghar@uha.fr, laurent.moalic@uha.fr and hojjat.rakhshani@uha.fr
Your application must contain the following documents:
A letter motivating the application (cover letter)
Curriculum vitae
Grade transcripts and BSc/MSc certificates
References
