1,720,966 research outputs found
Automated Neural Architecture Search for Cardiac Amyloidosis Classification from [18F]-Florbetaben PET Images
No full textMedical image classification using convolutional neural networks (CNNs) is promising but often requires extensive manual tuning for optimal model definition. Neural architecture search (NAS) automates this process, reducing human intervention significantly. This study applies NAS to [18F]-Florbetaben PET cardiac images for classifying cardiac amyloidosis (CA) sub-types (amyloid light chain (AL) and transthyretin amyloid (ATTR)) and controls. Following data preprocessing and augmentation, an evolutionary cell-based NAS approach with a fixed network macro-structure is employed, automatically deriving cells’ micro-structure. The algorithm is executed five times, evaluating 100 mutating architectures per run on an augmented dataset of 4048 images (originally 597), totaling 5000 architectures evaluated. The best network (NAS-Net) achieves 76.95% overall accuracy. K-fold analysis yields mean ± SD percentages of sensitivity, specificity, and accuracy on the test dataset: AL subjects (98.7 ± 2.9, 99.3 ± 1.1, 99.7 ± 0.7), ATTR-CA subjects (93.3 ± 7.8, 78.0 ± 2.9, 70.9 ± 3.7), and controls (35.8 ± 14.6, 77.1 ± 2.0, 96.7 ± 4.4). NAS-derived network performance rivals manually determined networks in the literature while using fewer parameters, validating its automatic approach’s efficacy
Analisi di immagini cerebrali 3D in Medicina Nucleare mediante modelli spiegabili di intelligenza artificiale (XAI)
No full textOverview delle applicazioni di Deep Learning nel campo delle Bioimmagini.
Creazione di una 3D CNN per la classificazione di immagini volumetriche PET F18-FDG dello stadio di demenza nella malattia di Alzheimer (CN, MCI, AD). Studio della variazione delle performance della rete applicando modifiche alla sua architettura ed effetti dell'applicazione di tecniche standard di Data Augmentation.
Applicazione di strumenti di Intelligenza Artificiale Spiegabile (XAI) volti a interpretare localmente la predizione della CNN (applicazione dell'algoritmo "LIME" a immagini volumetriche) e ad ispezionare le operazioni implementate layer per layer dalla rete mediante Mappe di Attivazione e Visualizzazione dei pesi dei filtri.
Overview of Deep Learning applications in Bioimaging.
Building of a 3D CNN for PET F18-FDG volumetric images classification of dementia stage in Alzheimer's disease (CN, MCI, AD). Variation of the network performance modifying its architecture and effects of the application of standard Data Augmentation techniques.
Application of Explainable Artificial Intelligence tools (XAI) for Local Interpretabilty of CNN prediction (application of the "LIME" algorithm to volumetric images) and inspection layer by layer of image elaboration visualizing the activations and first-layer weights
Explainable Deep Learning for Medical Imaging: Bridging the Gap Between AI Algorithms and Clinical Decision-Making
No full textNegli ultimi anni, l'Explainable Artificial Intelligence (XAI) sta assumendo un ruolo sempre più rilevante per migliorare l'interpretabilità dei modelli di Deep Learning (DL) nelle applicazioni di imaging medicale.
Questo interesse è alimentato dalle notevoli prestazioni ottenute dagli algoritmi di DL nell'esecuzione di una vasta gamma di compiti di analisi dell'imaging medico.
Il DL può significativamente supportare l'analisi dei dati medici sotto vari punti di vista, i quali includono la riduzione del carico di lavoro dei clinici, l'automatizzazione di compiti dispendiosi e garantendo prestazioni più elevate, maggiore oggettività e riproducibilità nelle valutazioni.
Il DL può inoltre servire come potenziale strumento di "knowledge-discovery'', in termini di estrazione di image-biomarkers ancora sconosciuti.
Tuttavia, la mancanza di trasparenza nella comprensione e spiegazione dei processi decisionali e dei fattori chiave che influenzano questi modelli solleva preoccupazioni etiche, di sicurezza e di affidabilità.
Questa limitazioni ne riducono l'accettabilità e creano scetticismo da parte degli utilizzatori, in particolare in settori ad alto rischio come la sanità.
La necessità di algoritmi trasparenti e spiegabili sta attirando anche l'attenzione delle autorità di regolamentazione, evidenziando che la ricerca XAI potrebbe influenzare direttamente lo sviluppo industriale di dispositivi software medici basati su AI.
Nonostante vi sia un crescente interesse in materia di XAI, non esiste un consenso ampio su come raggiungere una piena interpretabilità nei sistemi AI basati su DL.
Questa sfida è ulteriormente complicata dalla difficoltà di valutare i metodi XAI. Attualmente, non ci sono framework universalmente riconosciuti per valutare la qualità degli output prodotti da modelli di XAI, e la valutazione delle spiegazioni rimane un'area di ricerca complessa ed in evoluzione.
Questa tesi ha l'obiettivo di integrare la XAI nell'analisi dell'imaging medico per sviluppare sistemi di supporto decisionale che superino le limitazioni del paradigma del "black-box".
In recent years, Explainable Artificial Intelligence (XAI) has become increasingly relevant to enhance the interpretability of Deep Learning (DL) models in medical imaging applications.
This interest stems from the remarkable performance obtained by DL algorithms for the execution of a wide variety of relevant medical imaging analysis tasks.
DL can significantly aid in medical data analysis by reducing clinician workload, automating time-intensive tasks, and ensuring higher performance, objectivity, and reproducibility in evaluations. Additionally, DL holds potential as a knowledge-discovery tool, in terms of extracting yet unknown image-based biomarkers.
However, the lack of transparency in understanding and explaining the decision-making processes and key factors influencing these models raises ethical, safety, and trustworthiness concerns.
This limitation reduces their acceptance, particularly in high-risk areas such as healthcare.
The need for transparent and explainable algorithms is also drawing attention from regulatory bodies, highlighting that XAI research may directly influence the industrial development of AI-powered medical software devices.
However, despite this growing interest, there is no broad agreement on how to achieve full interpretability in DL-based AI systems.
This challenge is further compounded by the difficulty in evaluating XAI methods.
Currently, there are no universally recognized standard frameworks to assess the quality of XAI outputs, and the evaluation of explanations remains a complex and evolving research area.
This thesis aims to integrate XAI into medical imaging analysis to develop decision-support systems that overcome the limitations of the "black-box" paradigm
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Convolutional Neural Networks Latent-Space analysis with Reject Option: application to medical images classification
No full textDeep neural networks demonstrate performance on par with, or better than, clinicians in many tasks due to the rapid increase of available data and computing power. To comply with the principles of trustworthy AI, the AI system must be transparent, robust, fair, and ensure accountability. Therefore, ensuring the interpretability of deep neural networks (DNNs) might be crucial before they can be incorporated into the routine clinical workflow (Salahuddin et al., 2022).
Deep learning (DL) models are comprised of several layers which process and learn data representation across multiple levels of abstraction, without requiring human-engineered features. The multidimensional space generated through the process of feature extraction performed by the network’s layers might be referred to as latent space, or features space, and each dimension of the latent space corresponds to a specific feature or attribute that the DNN has identified within the input data. The nonlinearity and depth of these models, featuring tens or even hundreds of processing layers and thousands to millions of parameters, categorize them as black-box models with opaque internal mechanisms. Convolutional neural networks (CNNs) have emerged as the go-to standard for computer vision problems. Additionally, DL has showcased top-tier performance in numerous medical imaging challenges, especially concerning classification tasks (Jiang et al., 2023). Explainable Artificial Intelligence (XAI) refers to AI solutions that can provide insights into the internal workings of DL models in a manner understandable to the end-user. In the medical field, the purpose of AI is to assist physicians in performing their duties more efficiently and accurately, not to replace them. This collaboration requires trust from clinical experts, and trust is built on understanding (Lipton, 2017).
This thesis introduces and validates a novel reject option strategy designed for deep learning-based classifiers, demonstrating its applicability through implementation in three distinct case studies. The introduced reject option approach leverages Data Point Target Density (DPTD) for feature extraction from the latent space, a measure specifically devised and deployed for this purpose, alongside the k* value of the test sample under evaluation. To facilitate a deeper comprehension of the rejector's decisions, the CNN’s latent space in the last layer before the final classification was extracted and its dimensionality was condensed to a 2-D visualization using the t-SNE manifold learning technique. This visualization incorporated the individual Degree of Locality Preservation (DLP) value, providing insights into the fidelity of each data point's transition from the original high-dimensional space to the reduced embedding space
EXPLAINABLE ARTIFICIAL INTELLIGENCE IN MEDICAL IMAGES CLASSIFICATION BY PROTOTYPICAL PART NETWORK
No full textIn the medical imaging field, Convolutional Neural Networks (CNNs) present themselves as promising tools for diagnostic support due to their high performance. However, their opaque nature has slowed their adoption in healthcare. Explainable Artificial Intelligence (XAI) techniques aim to overcome this lack of transparency by enhancing interpretability without sacrificing performance. It is important to highlight, however, the need to identify reliable and standardized methods for evaluating the quality of explanations provided by these techniques.
Part-prototype models are a type of intrinsically explainable model that operate on images. These models integrate classification and explanation, which are based on the similarity between parts of a test image and prototypical parts (prototypes) of a specific class. The most representative implementation of this approach is the Prototypical Part Network (ProtoPNet), which leverages CNNs to learn prototypical parts of each class. In this work, we implement ProtoPNet to classify Normal/Pneumonia patients from chest X-ray images of a publicly available dataset, assessing the consistency and correctness of model’s explanations. We also investigate model’s generalization capability to unseen data.
ProtoPNet is composed by three modules: a CNN used as a feature extractor, a prototype layer and a fully connected layer which acts as decision layer. Using a specific loss function, the first two modules are trained to learn a meaningful latent space, in which prototypes of different classes are clustered in L2 distance around two separated centroids. During testing, the convolutional module compresses the input image into the latent space. Then, the prototype layer computes activation maps of similarities between the convolutional output and prototypes. These maps are max-pooled into similarity scores, which indicate how strongly a prototype is present in the input image. Such scores are finally weighted by the fully connected layer, producing class logits. The predicted class is the one with the highest logit. Additionally, the aforementioned activation maps are upsampled to visualize both prototypes and prototypical activations as image patches, making the model explainable both globally and locally.
The dataset chosen for this work contains 5856 frontal X-ray images of pediatric patients from a Chinese hospital. The dataset’s composition is 73% Pneumonia- 27% Normal. Pneumonia images include both bacterial and viral pneumonia. The former is usually reflected on the image with localized opacities, while the latter manifests with a more diffuse interstitial pattern in both lungs. In order to maintain dataset’s composition and to avoid data leakage we make a custom splitting of the dataset into training set (80%) and test set (20%).
Firstly, we train five ProtoPNets using 5-fold Cross Validation (CV) to assess whether the model’s accuracy and its explanations remain consistent across different dataset splits. By using a specific function, the splitting replicates dataset’s composition in all the folds and prevents data leakage. Images are converted to RGB, resized and transformed into tensors, who’s values are scaled into [0, 1]. Additionally, we standardize images with dataset’s mean and standard deviation for faster convergence. Since the dataset is unbalanced, we apply an offline data augmentation pipeline to training sets which includes geometric transformations and makes the two classes to have the same number of training images.
The consistency of the models’ explanations is assessed by examining both their global and local reproducibility. For this purpose, we introduce novel approaches. The Prototype average pair distance metric is utilized to monitor both the convergence of prototypes over epochs and their reproducibility in the latent space across folds (global explanation reproducibility). Additionally, we compute the L2 distance between inter-class prototype centroids in the latent space for each of the five models using hierarchical clustering. We then demonstrate that the aforementioned metric correlates with this distance by computing the Pearson linear correlation coefficient between the five metric values and the corresponding distances obtained from hierarchical clustering.
To quantify whether the five models provide consistent explanations for a given test image (local explanation reproducibility), we employ the Dice index to compute the average overlap between the most activated patches of the image by each model.
In the case of Part-prototype models, evaluating the correctness of models’ explanations translates to assessing the correctness of prototypical patches reconstruction. This was done through a single deletion experiment, which consists of deleting the prototypical patch from its source image and observing the change of the similarity score’s value when forwarding the perturbed image through the net. We also calculate, for the same image, the average of the similarity scores obtained by deleting random patches for comparison. To assess their generalization capability, we test all our 5-fold models with the hold-out internal test set. Furthermore, we evaluate the performance of our networks using images from an external dataset. This contains Normal and Pneumonia chest X-ray images from a more heterogeneous cohort of subjects in terms of age. In this case, the dataset is unbalanced in favor of the Normal class, and comprises images collected from an American hospital. We sample two test sets, one maintaining the composition of the internal dataset and the other reflecting that of the external dataset, named ’external 1’ and ’external 2’, respectively. The model that achieves the highest accuracy on the internal test set is chosen to present examples of local explanation of test images.
The 5-fold CV lead to similar validation accuracies across the folds (96.94 ± 0.34%), suggesting that the network’s performance is independent from data splitting. Regarding global explanation reproducibility, the proposed metric converges to the same value (44.14 ± 0.50) in all the five models and produces a high correlation coefficient (0.99) with the distances computed through hierarchical clustering. The local explanation reproducibility experiment shows that the models predominantly focus on the same patches to classify the same image, reflected by an average Dice index higher than 0.60 in 3 out of 4 examples. These results indicate that models trained on different data splits provide consistent explanations.
Results from the correctness experiment evidence that the deletion of the prototypical patch from its source image significantly decreases the similarity score, suggesting that patches reconstructed via upsampling are important for the network to identify a prototype in its source image. On the other hand, the random patch deletions unexpectedly lead to low similarity scores, so we cannot exclude the possibility that other pixels not included in the patches may also be important in representing the features learned by the prototype layer.
Our 5-fold models exhibit a mean accuracy of 97.27 ± 0.33% on the internal test set, which is close to the mean accuracy observed on the validation sets. The best model achieved an accuracy of 97.60%, which is comparable to the benchmark accuracy of 98.99% achieved by a black-box model. However, when tested on external datasets, their performance notably declines. The mean accuracies on external 1 and external 2 are 87.96 ± 2.08% and 78.84 ± 6.14%, respectively, indicating a significant decrease compared to the internal test set. These lower performances are particularly evident in the specificity metric, suggesting that our models are not able enough to recognize Normal images. Interestingly, the recall metric (sensitivity) remains relatively stable. This discrepancy in performance may be due to the impact of various factors when evaluating with external images, such as statistical differences between the two datasets and the imbalance of the dataset used for training. Local explanations of our best model show themselves as intuitive and self-explanatory. However, we deem a clinical evaluation of these explanations necessary.
In conclusion, our work demonstrates that ProtoPNet represents a promising explainable model, capable to maintain the performances of the black-box models. We have shown the consistency of model explanations through innovative approaches at both global and local levels. However, our results also indicate that further analysis have to be conducted to assess the correctness of ProtoPNet’s explanations. Moreover, results underline significant challenges in generalizing models to external data. Lastly, we deem a clinical evaluation of model explanations essential to ensure their utility and reliability in medical practice
Classificazione della malattia di Parkinson da immagini cerebrali SPECT attraverso reti neurali convoluzionali
No full textITA
La malattia di Parkinson è la seconda malattia neurodegenerativa più comune e il più frequente dei disordini del movimento. Si stima che tale malattia colpisca 4 milioni di persone in tutto il mondo: si osserva raramente nei pazienti di età inferiore ai 40 anni e l’incidenza aumenta con l’età. Tale malattia è meno frequente nel sesso femminile rispetto al maschile (40% femmine, 60% maschi) e in quest’ultimi l’esordio in media avviene due anni prima rispetto le donne. La diagnosi della malattia è di tipo clinico per cui, per stabilire criteri più oggettivi, accanto a tale valutazione acquista particolare rilevanza l’imaging diagnostico. Come supporto per il sospetto clinico del neurologo si ricorre ad esami eseguiti con PET (Tomografia a emissione di positroni), fMRI (Risonanza Magnetica Funzionale) e SPECT (Tomografia a emissione di fotone singolo). In particolare, la SPECT con FP-CIT è indicata in tutte quelle situazioni in cui, per il neurologo, è importante sapere se c’è perdita delle terminazioni dopaminergiche nello striato e assume un ruolo determinante nella diagnosi di Parkinsonismo per tutti quei pazienti, valutati in fase precoce, il cui quadro clinico non è ancora delineato visto il lento e graduale esordio dei sintomi. Tali sintomi si manifestano quando la ‘substanzia nigra’ ha perso circa il 60% dei neuroni dopaminergici e la dopamina residua è l’80% rispetto ai valori normali. Ad oggi non esiste una cura per la malattia di Parkinson, tuttavia, esistono alcuni trattamenti che possono alleviarne i sintomi e migliorare la qualità della vita del paziente. Mentre la malattia di Parkinson presenta un quadro clinico ben definito caratterizzato da sintomi motori come tremori, rigidità e bradicinesia e, non motori come declino cognitivo, disautonomia e disturbi del sonno, i Parkinsonismi, invece, includono un insieme più ampio di condizioni che possono manifestare sintomi simili ma con cause differenti. A tal proposito è importante una diagnosi corretta fin dai primi esordi della malattia in modo da applicare un approccio terapeutico adeguato. L’obiettivo della tesi è quindi implementare un modello di rete neurale convoluzionale (CNN) per la classificazione tra la malattia di Parkinson e i vari Parkinsonismi basandosi su volumi DICOM derivati da esami SPECT FP-CIT. La rete proposta rappresenta un modello di apprendimento supervisionato il cui dataset è costituito dai volumi di partenza e le relative label, ovvero il tipo di patologia. Sono state definite due varianti di modello, una per la classificazione Binaria tra pazienti con SPECT Positiva e SPECT Negativa e l’altra per la classificazione Multiclasse tra pazienti con malattia di Parkinson, Tremore Essenziale e non patologici. Questo approccio mira a fornire un mezzo diagnostico più efficiente e preciso, che potrebbe essere risolutivo soprattutto in situazioni in cui la diagnosi clinica è incerta o ambigua, contribuendo così ad una diagnosi più tempestiva e ad un trattamento mirato.
ENG
Parkinson's disease is the second most common neurodegenerative disease and the most frequent among movement disorders. It’s estimated that this disease affects 4 million people worldwide: it is rarely observed in patients under the age of 40, and its incidence increases with age. The disease is less common in females compared to males (40% females, 60% males), and in males, onset typically occurs two years earlier than in females. The diagnosis of the disease is clinical, but to establish more objective criteria, diagnostic imaging becomes particularly relevant alongside this evaluation. To support the neurologist's clinical suspicion, examinations performed with PET (Positron Emission Tomography), fMRI (Functional Magnetic Resonance Imaging), and SPECT (Single Photon Emission Computed Tomography) are used. In particular, SPECT with FP-CIT is indicated in all those situations where it is important for the neurologist to know whether there is a loss of dopaminergic terminations in the striatum, and it plays a crucial role in the diagnosis of Parkinsonism for those patients evaluated in the early phase whose clinical picture is not yet delineated due to the slow and gradual onset of symptoms. These symptoms manifest when the 'substantia nigra' has lost about 60% of dopaminergic neurons, and the residual dopamine is 80% compared to normal values. To date, there is no cure for Parkinson's disease; however, there are some treatments that can alleviate symptoms and improve the patient's quality of life. While Parkinson’s disease present a well-defined clinical picture characterized by motor symptoms such as tremors, rigidity, and bradykinesia, as well as non-motor symptoms such as cognitive decline, autonomic dysfunction, and sleep disturbances, Parkinsonisms, on the other hand, include a broader range of conditions that may manifest similar symptoms but with different causes. In this regard, a correct diagnosis from the early stages of the disease is important in order to apply an appropriate therapeutic approach. The objective of the thesis is therefore to implement a convolutional neural network (CNN) model for the classification between Parkinson's disease and various Parkinsonisms based on DICOM volumes derived from FP-CIT SPECT examinations. The proposed network represents a supervised learning model whose dataset consists of input volumes and their respective labels, i.e., the type of pathology. Two model variants have been defined, one for Binary classification between patients with Positive and Negative SPECT and the other for Multiclass classification between patients with Parkinson's disease, Essential Tremor, and non-pathological cases. This approach aims to provide a more efficient and precise diagnostic tool, which could be decisive especially in situations where the clinical diagnosis is uncertain or ambiguous, thus contributing to a more timely diagnosis and targeted treatment
SVILUPPO DI ALGORITMI DI DEEP LEARNING PER LA DIAGNOSI DELLA MALATTIA DI PARKINSON DA IMMAGINI SPECT E DATI CLINICI
No full textSviluppo di algoritmi di deep learning per la diagnosi della malattia di Parkinson da immagini SPECT e dati clinici
Development of Deep Learning Algorithms for the Diagnosis of Parkinson's Disease from SPECT Images and Clinical Dat
- …
