Fabrizio Nunnari - Academia.edu (original) (raw)

Papers by Fabrizio Nunnari

2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW)

In recent times, information presentation has evolved towards sophisticated approaches that invol... more In recent times, information presentation has evolved towards sophisticated approaches that involve multi-modal aspects and character-based mediation. This paper presents a novel methodology for creating information presentations based on a dramatization of the content exposition in two respects. On one side, the author plots a character's monologue that aims at achieving presentation goal and exhibits an engaging inner conflict; on the other side, the system architecture dynamically assembles the elementary units of the plot scripted by the author by implementing a tension between contrasting communicative functions. The methodology has been applied in the implementation of a virtual guide to an historical site.

Until now, Coronavirus SARS-CoV-2 has caused more than 850,000 deaths and infected more than 27 m... more Until now, Coronavirus SARS-CoV-2 has caused more than 850,000 deaths and infected more than 27 million individuals in over 120 countries. Besides principal polymerase chain reaction (PCR) tests, automatically identifying positive samples based on computed tomography (CT) scans can present a promising option in the early diagnosis of COVID-19. Recently, there have been increasing efforts to utilize deep networks for COVID-19 diagnosis based on CT scans. While these approaches mostly focus on introducing novel architectures, transfer learning techniques, or construction large scale data, we propose a novel strategy to improve the performance of several baselines by leveraging multiple useful information sources relevant to doctors' judgments. Specifically, infected regions and heat maps extracted from learned networks are integrated with the global image via an attention mechanism during the learning process. This procedure not only makes our system more robust to noise but also guides the network focusing on local lesion areas. Extensive experiments illustrate the superior performance of our approach compared to recent baselines. Furthermore, our learned network guidance presents an explainable feature to doctors as we can understand the connection between input and output in a grey-box model.

Lecture Notes in Computer Science, 2021

Dermatologists recognize melanomas by inspecting images in which they identify human-comprehensib... more Dermatologists recognize melanomas by inspecting images in which they identify human-comprehensible visual features. In this paper, we investigate to what extent such features correspond to the saliency areas identified on CNNs trained for classification. Our experiments, conducted on two neural architectures characterized by different depth and different resolution of the last convolutional layer, quantify to what extent thresholded Grad-CAM saliency maps can be used to identify visual features of skin cancer. We found that the best threshold value, i.e., the threshold at which we can measure the highest Jaccard index, varies significantly among features; ranging from 0.3 to 0.7. In addition, we measured Jaccard indices as high as 0.143, which is almost 50% of the performance of state-of-the-art architectures specialized in feature mask prediction at pixel-level, such as U-Net. Finally, a breakdown test between malignancy and classification correctness shows that higher resolution saliency maps could help doctors in spotting wrong classifications.

The classification of skin lesion images is known to be biased by artifacts of the surrounding sk... more The classification of skin lesion images is known to be biased by artifacts of the surrounding skin, but it is still not clear to what extent masking out healthy skin pixels influences classification performances, and why. To better understand this phenomenon, we apply different strategies of image masking (rectangular masks, circular masks, full masking, and image cropping) to three datasets of skin lesion images (ISIC2016, ISIC2018, and MedNode). We train CNN-based classifiers, provide performance metrics through a 10-fold cross-validation, and analyse the behaviour of Grad-CAM saliency maps through an automated visual inspection. Our experiments show that cropping is the best strategy to maintain classification performance and to significantly reduce training times as well. Our analysis through visual inspection shows that CNNs have the tendency to focus on pixels of healthy skin when no malignant features can be identified. This suggests that CNNs have the tendency of “eagerly” ...

ArXiv, 2020

A shortage of dermatologists causes long wait times for patients who seek dermatologic care. In a... more A shortage of dermatologists causes long wait times for patients who seek dermatologic care. In addition, the diagnostic accuracy of general practitioners has been reported to be lower than the accuracy of artificial intelligence software. This article describes the Skincare project (H2020, EIT Digital). Contributions include enabling technology for clinical decision support based on interactive machine learning (IML), a reference architecture towards a Digital European Healthcare Infrastructure (also cf. EIT MCPS), technical components for aggregating digitised patient information, and the integration of decision support technology into clinical test-bed environments. However, the main contribution is a diagnostic and decision support system in dermatology for patients and doctors, an interactive deep learning system for differential diagnosis of malignant skin lesions. In this article, we describe its functionalities and the user interfaces to facilitate machine learning from huma...

ArXiv, 2019

We describe a software toolbox for the configuration of deep neural networks in the domain of ski... more We describe a software toolbox for the configuration of deep neural networks in the domain of skin cancer classification. The implemented software architecture allows developers to quickly set up new convolutional neural network (CNN) architectures and hyper-parameter configurations. At the same time, the user interface, manageable as a simple spreadsheet, allows non-technical users to explore different configuration settings that need to be explored when switching to different data sets. In future versions, meta leaning frameworks can be added, or AutoML systems that continuously improve over time. Preliminary results, conducted with two CNNs in the context melanoma detection on dermoscopic images, quantify the impact of image augmentation, image resolution, and rescaling filter on the overall detection performance and training time.

User Modeling and User-Adapted Interaction, 2008

In this paper, we present an application for character-based guided tours on mobile devices. The ... more In this paper, we present an application for character-based guided tours on mobile devices. The application is based on the Dramatour methodology for information presentation, which incorporates a dramatic attitude in character-based presentations. The application has been developed for a historical site and is based on a virtual character, "Carletto", a spider with an anthropomorphic aspect, who engages in a dramatized presentation of the site. Content items are delivered in a location-aware fashion, relying on a wireless network infrastructure, with visitors who can stroll freely. The selection of contents keeps track of user location and of the interaction history, in order to deliver the appropriate type and quantity of informative items, and to manage the given/new distinction in discourse. The communicative strategy of the character is designed to keep it believable along the interaction with the user, while enforcing dramatization effects. The design of the communicative strategy relies on

The15th International Conference on PErvasive Technologies Related to Assistive Environments

Companion Publication of the 2021 International Conference on Multimodal Interaction, 2021

Springer International Publishing eBooks, Aug 17, 2021

This paper presents an investigation on the task of anomaly detection for images of skin lesions.... more This paper presents an investigation on the task of anomaly detection for images of skin lesions. The goal is to provide a decision support system with an extra filtering layer to inform users if a classifier should not be used for a given sample. We tested anomaly detectors based on autoencoders and three discrimination methods: feature vector distance, replicator neural networks, and support vector data description fine-tuning. Results show that neural-based detectors can perfectly discriminate between skin lesions and open world images, but class discrimination cannot easily be accomplished and requires further investigation.

Transpilation allows to write code once and re-use it across multiple runtime environments. In th... more Transpilation allows to write code once and re-use it across multiple runtime environments. In this paper, we propose a software development practice to implement once the motion controllers of virtual humans and re-use the implementation in multiple game engines. In a case study, three common human behaviors – blinking, text-to-speech, and eye-gaze – were developed in the Haxe programming language and deployed in the free, open-source Blender Game Engine and the commercial Unity engine. Performance tests show that transpiled code executes within 67% faster to 127% slower with respect to an implementation manually written in the game engine target languages.

ArXiv, 2021

ELLÁK SOMFAI∗, Eötvös University, Hungary and Wigner Research Centre for Physics, Hungary BENJÁMI... more ELLÁK SOMFAI∗, Eötvös University, Hungary and Wigner Research Centre for Physics, Hungary BENJÁMIN BAFFY, Eötvös University, Hungary KRISTIAN FENECH, Eötvös University, Hungary CHANGLU GUO, Eötvös University, Hungary RITA HOSSZÚ, Semmelweis University, Hungary DORINA KORÓZS, Semmelweis University, Hungary MARCELL PÓLIK, Eötvös University, Hungary ATTILA ULBERT, Eötvös University, Hungary ANDRÁS LŐRINCZ, Eötvös University, Hungary

AVI provided an attractive opportunity for exploring novel visual interfaces for cultural heritag... more AVI provided an attractive opportunity for exploring novel visual interfaces for cultural heritage (CH). CH traditionally draws a lot of research attention when it comes to exploring the potential benefits from application of novel technology in realistic settings. At the same time, AVI focusses on exploring the state of the art visual interfaces and their application in various domains. The AVI-CH workshop nicely demonstrated the potential of combining these two aspects-the state of the art interfaces technologies with the information rich CH domain. The result was a number of high-quality submissions, with the diversity of topics presented by the papers accepted and discussed at the workshop.

Virtual Reality and Augmented Reality, 2019

Within the BRAVO project, we are designing four virtual characters that will interact with childr... more Within the BRAVO project, we are designing four virtual characters that will interact with children affected by ADHD. In order to assess the quality of the designed characters, we propose a metric to subjectively evaluate the level of intelligibility of the character’s facial expression. The results of a preliminary user study conducted with 23 individuals show that our quality measure can be used to quickly identify flawed expressions and iteratively improve the design of the characters.

KI 2021: Advances in Artificial Intelligence, 2021