Auteur : Support

Du 1er au 19 Juillet 2019, Université de Yaoundé I, Yaoundé, Cameroun.

Détails : http://ema.cri-info.cm/index.php/page-d-exemple/

Du 28 au 31 Octobre 2019, Rome, Italie. voir le pdf

AI methodologies to support automated analysis of sight threatening eye diseases

Sight threatening diseases, such as diabetic retinopathy and age related maculardegeneration have contributed to the 40% increase in outpatient attendances inthe last decade but are amenable to early detection and monitoring.Age-related macular degeneration (AMD) is a degenerative retinal disease thatcan cause irreversible vision loss. AMD affect 1 in 7 over the age of 50. It is theleading cause of blindness in Europe and North America and accounts for overhalf of partially sighted or legally blind certifications in the UK. Diabeticretinopathy (DR) is another leading cause of blindness in working agepopulations in the developed world. It is estimated that up to 50% of peoplewith proliferative DR who do not receive timely treatment will become legallyblind within 5 years. Up to 98% of severe visual loss due to DR can beprevented with early detection and treatment. Two imaging modalities, namely,color fundus photograph and optical coherence tomography are typically usedto diagnose these disease and to monitor their progression. Manual assessmentof these images is time consuming and cost intensive, which is also subjectiveand infeasible in many circumstances. Artificial intelligent methods areeffective alternative.He will discuss about several machine learning and/or deep learning approachesdeveloped by this team to facilitate automated analysis of AMD and DR

Skin Lesion Features Extraction and Classification with SVM in Real Numbers Clifford Algebra and Neural Network

The speech will start with the introduction of methods and techniques forfeatures extraction from skin lesion images. The features under considerationare from melanoma diagnosing systems in clinical use. Next, the speaker willshow features selection and ranking according to the features significance forskin lesion classification. This way a subset of 5 features is determined as mostimportant and they are used to create a skin lesion signature in the form of skinlesion feature vector (LFV). Further, the speaker will present binary, ternaryand quaternary support vector machines (SVMs) to classify a skin lesion to twoclasses (benign-B and melanoma-M), three classes (B, dysplastic nevi-D, andM), four classes (B, dysplastic nevi mild- Dm, dysplastic nevi severe – DS, andM). During the next part of the talk the presenter will acquaint the audiencewith basic concepts from Clifford Algebras: Clifford product and calculus,multi-vector, Clifford sub-spaces and operators. Then he will show thedevelopment of new formulae for calculation of the multi-vector coefficientsand their implementation to map a multi-vector to Clifford Algebra sub-spaces.Then the speaker will show skin lesion classification results by SVM in theClifford Algebra sub-spaces. On the next stage he will compare the obtainedresults with classification results: in the field of real numbers; diagnosing rulesin clinical practice; and contemporary neural networks (NN). The speech will end with a brief description a convolutional NN architecture and itsimplementation with a stochastic gradient descent learning classification of a3800 images from the ISIC2017 dataset.He will discuss about several machine learning and/or deep learning approachesdeveloped by this team to facilitate automated analysis of AMD and DR.