Comunicacions a congressos (Matemàtiques i Informàtica)

URI permanent per a aquesta col·leccióhttps://diposit.ub.edu/handle/2445/54812

Estadístiques

Examinar

Mostrant 1 - 16 de 16

From hand-crafted radiomics to deep learning: evaluating breast cancer classification methods in mammograms
(SPIE, 2025-04-10) Guzman Requena, Alejandro; Márquez Vara, Noah; Díaz, Oliver
This study evaluates the performance of some machine learning (ML) and deep learning (DL) models for breast cancer tumor classification in mammography (MG) images, by training them on the BCDR dataset. The study compares the use of radiomics-based features in ML models, including Random Forest, Support Vector Machines, and XGBoost, with two deep learning approaches using Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). Radiomics features were extracted from segmented regions of interest (ROIs) and used to train the ML models, performing hyperparameter tuning and cross-validation to optimize the results. CNN and ViT models were trained using the MG location present in the ROI segmentations, to explore the impact of tumor region localization assistance on classification performance. To examine and verify the performance of the ML models and the ViT, the area under the receiver operating characteristic curve (AUC-ROC) and the training execution time of all experiments (performed on the same device) were used. The results indicate that, while all methods achieve good performance on the training dataset (mean AUCROC scores around 0.9), they exhibit substantial performance drops when tested on external data. Among the evaluated models, ViT achieves the highest overall AUC-ROC in both internal (0.93) and external (0.68) validation, surpassing CNNs and radiomics-based ML models. However, ViT also incurs the highest computational cost, highlighting a trade-off between accuracy and training time. These findings underscore the need for multicenter, multi-vendor data to improve model generalization and reliability, as well as for continued refinement of advanced architectures, such as transformers, to optimize breast cancer lesion classification in clinical settings.
Enhancing the Utility of Privacy-Preserving Cancer Classification using Synthetic Data
(Springer, 2025-02-11) Osuala, Richard; Lang, Daniel M.; Riess, Anneliese; Kaissis, Georgios; Szafranowska, Zuzanna; Skorupko, Grzegorz; Díaz, Oliver; Schnabel, Julia A.; Lekadir, Karim, 1977-
Deep learning holds immense promise for aiding radiologists in breast cancer detection. However, achieving optimal model performance is hampered by limitations in availability and sharing of data commonly associated to patient privacy concerns. Such concerns are further exacerbated, as traditional deep learning models can inadvertently leak sensitive training information. This work addresses these challenges exploring and quantifying the utility of privacy-preserving deep learning techniques, concretely, (i) differentially private stochastic gradient descent (DP-SGD) and (ii) fully synthetic training data generated by our proposed malignancy-conditioned generative adversarial network. We assess these methods via downstream malignancy classification of mammography masses using a transformer model. Our experimental results depict that synthetic data augmentation can improve privacy-utility tradeoffs in differentially private model training. Further, model pretraining on synthetic data achieves remarkable performance, which can be further increased with DP-SGD fine-tuning across all privacy guarantees. With this first in-depth exploration of privacy-preserving deep learning in breast imaging, we address current and emerging clinical privacy requirements and pave the way towards the adoption of private high-utility deep diagnostic models. Our reproducible codebase is publicly available at https://github.com/RichardObi/mammo_dp.
Mitigating Overdiagnosis Bias in CNN-Based Alzheimer’s Disease Diagnosis for the Elderly
(2024-10-13) Ngoc Dang, Vien; Casamitjana, Adrià; Hernández-González, Jerónimo; Lekadir, Karim, 1977-
Diagnosing Alzheimer’s disease (AD) presents significant challenges in the oldest populations due to overlapping symptoms of normal cognitive aging and early-stage dementia. While AI algorithms have matched specialist performance in diagnosing AD, they tend to produce unreliable results for the oldest populations, generating false positives that increase radiologist workloads and healthcare costs. In this study, we focus on mitigating overdiagnosis bias in CNN-based AD diagnosis for these groups. We present a post-hoc bias mitigation technique that significantly improves fairness by reducing overdiagnosis and enhances reliability by improving calibration without compromising overall model accuracy. Code is available at: https://github.com/ngoc-vien-dang/C-GTOP.
Fat-suppressed breast MRI synthesis for domain adaptation in tumour segmentation
(2025-02-12) Garrucho, Lidia; Delegue, Eve; Osuala, Richard; Kessler, Dimitri; Kushibar, Kaisar; Díaz, Oliver; Lekadir, Karim, 1977-; Igual Muñoz, Laura
Heterogeneity in dynamic contrast-enhanced breast MRI acquisition protocols hinders the generalization of automatic tumour segmentation tools. While fat-suppressed MRI acquisition is common, some vendors do not provide these sequences, making a segmentation model trained with fat-suppressed images unusable for non-fat-suppressed cases. In this study, we propose two strategies to alleviate this issue. The first approach involves translating non-fat-suppressed to fat-suppressed breast MRI. The second approach integrates synthetic non-fat-suppressed MRI into the training pipeline of tumour segmentation models. Our experimental results demonstrate that both approaches significantly improve segmentation performance on non-fat-suppressed MRI, suggesting that domain adaptation techniques based on image synthesis can enhance the accuracy and reliability of tumour segmentation in breast MRI. The generative models will be made publicly available at medigan library (medigan [18] GitHub repository).
From the clinic: A survey on trustworthy AI in breast cancer
(IEEE, 2023) Joshi, Smriti; Emelie, Anais; Bobowicz, Maciej; Tsakou, Gianna; Charalambous, Stefanie; Salahuddin, Zohaib; Díaz, Oliver; Lekadir, Karim, 1977-
With the fast-growing applications of artificial intelligence (AI) in healthcare, it is essential to keep track of their credibility and reliability. We conducted a survey with healthcare practitioners to obtain requirements for developing reliable AI tools for breast cancer. We share our findings with the healthcare community, hoping that our work serves as a resource to build extensively validated and trustworthy solutions.
Sharing generative models instead of private data: a simulation study on mammography patch classification
(SPIE, 2022) Szafranowska, Zuzanna; Osuala, Richard; Breier, Bennet; Kushibar, Kaisar; Lekadir, Karim, 1977-; Díaz, Oliver
Early detection of breast cancer in mammography screening via deep-learning based computer-aided detection systems shows promising potential in improving the curability and mortality rates of breast cancer. However, many clinical centres are restricted in the amount and heterogeneity of available data to train such models to (i) achieve promising performance and to (ii) generalise well across acquisition protocols and domains. As sharing data between centres is restricted due to patient privacy concerns, we propose a potential solution: sharing trained generative models between centres as substitute for real patient data. In this work, we use three well known mammography datasets to simulate three different centres, where one centre receives the trained generator of Generative Adversarial Networks (GANs) from the two remaining centres in order to augment the size and heterogeneity of its training dataset. We evaluate the utility of this approach on mammography patch classification on the test set of the GAN-receiving centre using two different classification models, (a) a convolutional neural network and (b) a transformer neural network. Our experiments demonstrate that shared GANs notably increase the performance of both transformer and convolutional classification models and highlight this approach as a viable alternative to inter-centre data sharing. Find our code at https://github.com/ zuzaanto/mammo_gans_iwbi2022
Breast composition measurements from Full-Field Digital Mammograms using generative adversarial networks
(2024) García Marcos, Eloy; Badó Llardera, Xavier; Mann, Ritse M.; Osuala, Richard; Martí Marly, Robert
Breast density has demonstrated to be an important risk factor for the development of breast cancer and, therefore, different fully automated density assessment tools have been introduced to obtain quantitative glandu- lar tissue measures. Density maps (DMs) provide local tissue information, representing the amount of glandular tissue between the image receptor and the x-ray source at every pixel in the image. Usually, DMs are obtained from for processing, i.e. raw, mammograms. This fact could become a tricky problem because this type of images are not preserved in the clinical setting. The aim of this work is to introduce a deep learning based framework to synthesize glandular tissue DMs from for presentation mammograms. First, the breast region is located using a dedicated object detector network. Next, a generative adversarial network is used to obtain synthetic density maps, that are useful to evaluate not only the glandular tissue distribution but also the total glandular tissue volume within the breast. Results show that synthetic DMs obtain a structural similarity index of SSIM = 0.93 ± 0.06 with respect to real images. Similarly, shared information between the real and syn- thetic images, computed using the histogram intersection, corresponds to HI = 0.84 ± 0.10, while the average pixel difference represents only 3.85 ± 2.78 % of breast thickness. Furthermore, glandular tissue volume (GTV) obtained from synthetic density map show a strong correlation with the value provided by the real one (ρ = 0.89 [C.I 0.87 − 0.91]). In conclusion, generative deep learning models can be useful to evaluate breast composition, from local to global tissue distribution.
Pre- to post-contrast breast MRI synthesis for enhanced tumour segmentation
(SPIE, 2024) Osuala, Richard; Joshi, Smriti; Tsirikoglou, Apostolia; Garrucho, Lidia; López Pinaya, Walter Hugo; Díaz, Oliver; Lekadir, Karim, 1977-
Despite its benefits for tumour detection and treatment, the administration of contrast agents in dynamic contrast-enhanced MRI (DCE-MRI) is associated with a range of issues, including their invasiveness, bioaccu- mulation, and a risk of nephrogenic systemic fibrosis. This study explores the feasibility of producing synthetic contrast enhancements by translating pre-contrast T1-weighted fat-saturated breast MRI to their corresponding first DCE-MRI sequence leveraging the capabilities of a generative adversarial network (GAN). Additionally, we introduce a Scaled Aggregate Measure (SAMe) designed for quantitatively evaluating the quality of synthetic data in a principled manner and serving as a basis for selecting the optimal generative model. We assess the generated DCE-MRI data using quantitative image quality metrics and apply them to the downstream task of 3D breast tumour segmentation. Our results highlight the potential of post-contrast DCE-MRI synthesis in enhancing the robustness of breast tumour segmentation models via data augmentation. Our code is available at https://github.com/RichardObi/pre_post_synthesis.
Mitigating annotation shift in cancer classification using single image generative models
(SPIE, 2024) Buetas Arcas, Marta; Osuala, Richard; Lekadir, Karim, 1977-; Díaz, Oliver
Artificial Intelligence (AI) has emerged as a valuable tool for assisting radiologists in breast cancer detection and diagnosis. However, the success of AI applications in this domain is restricted by the quantity and quality of available data, posing challenges due to limited and costly data annotation procedures that often lead to annotation shifts. This study simulates, analyses and mitigates annotation shifts in cancer classification in the breast mammography domain. First, a high-accuracy cancer risk prediction model is developed, which effectively distinguishes benign from malignant lesions. Next, model performance is used to quantify the impact of annotation shift. We uncover a substantial impact of annotation shift on multiclass classification performance particularly for malignant lesions. We thus propose a training data augmentation approach based on single-image generative models for the affected class, requiring as few as four in-domain annotations to considerably mitigate annotation shift, while also addressing dataset imbalance. Lastly, we further increase performance by proposing and validating an ensemble architecture based on multiple models trained under different data augmentation regimes. Our study offers key insights into annotation shift in deep learning breast cancer classification and explores the potential of single-image generative models to overcome domain shift challenges. All code used for this study is made publicly available at https://github.com/MartaBuetas/EnhancingBreastCancerDiagnosis.
A study on the role of radiomics feature stability in predicting breast cancer subtypes
(SPIE, 2024) Cama, Isabella; Guzman Requena, Alejandro; Garbarino, Sara; Campi, Cristina; Lekadir, Karim, 1977-; Díaz, Oliver
Imaging features (radiomics) have potential for predicting Triple Negative Breast Cancer and other subtypes using magnetic resonance images (MRI). This work uses 244 images from the Duke-Breast-Cancer-MRI dataset to investigate the complex interplay between radiomics feature stability, with respect to segmentation variability, and prediction results of machine learning models. Our analysis reveals that features demonstrating high stability across different segmentations tend to enhance model performance, whereas unstable features sensitive to small segmentation changes degrade predictive accuracy. This exploration underscores the importance of feature stability in the development of reliable models for breast cancer subtype classification.
Leveraging epistemic uncertainty to improve tumour segmentation in breast MRI: an exploratory analysis
(SPIE, 2024) Joshi, Smriti; Osuala, Richard; Garrucho, Lidia; Tsirikoglou, Apostolia; Riego, Javier del; Gwoździewicz, Katarzyna; Kushibar, Kaisar; Díaz, Oliver; Lekadir, Karim, 1977-
Medical image segmentation has improved with deep-learning methods, especially for tumor segmentation. However, variability in tumor shapes, sizes, and enhancement remains a challenge. Breast MRI adds further uncertainty due to anatomical differences. Informing clinicians about result reliability and using model uncertainty to improve predictions are essential. We study Monte-Carlo Dropout for generating multiple predictions and finding consensus segmentation. Our approach reduces false positives using per-pixel uncertainty and improves segmentation metrics. In addition, we study the correlation of model performance to the perceived ease of manual segmentation. Finally, we compare the per-pixel uncertainty with the inter-rater variability as segmented by six different radiologists. Our code is available at https://github.com/smriti-joshi/uncertainty-segmentation-mcdropout.git.
Multi-Objective Reinforcement Learning for Designing Ethical Environments
(International Joint Conferences on Artificial Intelligence, 2021) Rodríguez Soto, Manel; López Sánchez, Maite; Rodríguez-Aguilar, Juan A. (Juan Antonio)
AI research is being challenged with ensuring that autonomous agents learn to behave ethically, namely in alignment with moral values. A common approach, founded on the exploitation of Reinforcement Learning techniques, is to design environments that incentivise agents to behave ethically. However, to the best of our knowledge, current approaches do not theoretically guarantee that an agent will learn to behave ethically. Here, we make headway along this direction by proposing a novel way of designing environments wherein it is formally guaranteed that an agent learns to behave ethically while pursuing its individual objectives. Our theoretical results develop within the formal framework of Multi-Objective Reinforcement Learning to ease the handling of an agent's individual and ethical objectives. As a further contribution, we leverage on our theoretical results to introduce an algorithm that automates the design of ethical environments.
Avaluació de l'accessibilitat per a persones amb baixa visió dels gràfics estadístics dels llocs web de les universitats públiques catalanes
(Universitat de Barcelona, 2020-03-28) Alcaraz Martínez, Rubén; Ribera, Mireia; Granollers Saltiveri, Toni
Els gràfics estadístics permeten visualitzar dades d'una manera més eficient que altres formats, com ara les taules. El sistema educatiu considera la necessitat que els estudiants assoleixin competències relacionades amb la interpretació i la creació de gràfics. La recerca i el periodisme són altres àmbits on els gràfics tenen una presència important i que justifiquen la necessitat de fer-los accessibles per garantir l'accés de les persones discapacitades a aquests sectors clau de la societat. Els processos d'avaluació de la qualitat de les titulacions universitàries han significat la posada en obert d'un conjunt d'indicadors en forma de gràfics en els webs de les universitats: estadístiques que orienten els estudiants sobre quin centre i quin ensenyament poden triar. En el període 2012-2016 el nombre d'estudiants discapacitats matriculats a les universitats catalanes s'ha incrementat un 1.341,67 %. La bibliografia recull diferents treballs centrats en l'accessibilitat de les persones cegues o amb poca resta de visió, als gràfics, però no aborda el cas específic dels usuaris amb baixa visió, fet que contribueix a marginar un col·lectiu que representa el 97 % de les persones discapacitades visuals del món. Darrerament, a Europa i Espanya s'han realitzat importants avenços legislatius en matèria d'accessibilitat informàtica, com la Directiva (UE) 2019/882 i el Reial decret 1112/2018. Tanmateix, encara queda molt marge de millora quant al compliment de les normatives aplicables en el context universitari. L'accessibilitat dels gràfics estadístics per a persones amb baixa visió implica diferents consideracions no abordades per les directrius internacionals, que se centren fonamentalment en les persones cegues i en l'ús d'alternatives textuals o taules, solucions amb les quals l'usuari amb baixa visió perd la possibilitat de gaudir dels beneficis de la visualització pel que fa a l'eficiència en la percepció i comprensió de les dades. Metodologia Amb l'objectiu de determinar el nivell d'accessibilitat dels gràfics estadístics dels webs de les universitats, s'ha desenvolupat una eina d'avaluació específica basada en heurístiques. La mostra està formada per dues universitats públiques de diferents magnituds (UB i UdL) i pels webs EUC i Winddat de l'AQU, llocs de referència per a moltes universitats que enllacen amb els seus gràfics. De cada web s'ha analitzat una pàgina representativa dins dels apartats d'informació sobre els processos del MVSMA. La unitat de valoració considerada és la pàgina completa. En el cas de pàgines amb diversos gràfics, s'han analitzat tots i s'ha realitzat la mitjana de les puntuacions obtingudes. La recollida de dades s'ha realitzat durant el mes de setembre de 2019. Resultats i treball futur Cap dels webs analitzats presenta gràfics que es puguin considerar accessibles per al col·lectiu objecte d'estudi. Els principals problemes són l'ús del color, l'accés a través del teclat i l'absència d'alternatives textuals. Com a principal línia de treball, una vegada consolidada la metodologia, s'ampliarà la mostra per obtenir un resultat més representatiu de l'estat de l'accessibilitat dels gràfics estadístics als webs de les universitats. La metodologia serà aplicable a altres àmbits, com la docència o la recerca. També es treballa per proposar una guia per a l'elaboració d'aquest tipus de contingut.
Towards Global Explanations for Credit Risk Scoring
(Neural Information Processing Systems Foundation, 2018-11-23) Unceta, Irene; Nin, Jordi; Pujol Vila, Oriol
In this paper we propose a method to obtain global explanations for trained black-box classifiers by sampling their decision function to learn alternative interpretable models. The envisaged approach provides a unified solution to approximate non-linear decision boundaries with simpler classifiers while retaining the original classification accuracy. We use a private residential mortgage default dataset as a use case to illustrate the feasibility of this approach to ensure the decomposability of attributes during pre-processing.
Preferències dels estudiants en relació al tema d’estudi del TFG de Farmàcia (UB)
(Universitat de Barcelona, 2015-02) Aróztegui Trenchs, Montserrat; Bernal Serrano, Antonio; Bonet Clols, Francesc; Bosque Pueyo, Ramón; Cambras Riu, Trinitat; Canudas Teixidó, Anna-Maria; Domínguez García, Àngela; Engel Rocamora, Pablo; Escribano Ferrer, Elvira; Fernández Lastra, Cecilia; Folch Sánchez, Montserrat; López Sabater, María del Carmen; March Pujol, Marian; Marqués Villavecchia, Ana M.; Marrero González, Pedro F.; Miquel Colomé, Jordi; Muñoz Juncosa, Montserrat; Palazón Barandela, Javier; Piqué Benages, Maria Esther; Puignou i Garcia, Lluís; Pujol Dilmé, M. Dolors; Simon Pallisé, Joan; Tebar Ramon, Francesc; Ticó Grau, Josep R.; Escubedo Rafa, Elena; Zulaica Gallego, Ester
El TFG del grau de farmàcia UB es porta a terme en el marc d’un àmbit docent principal i integra coneixements de com a mínim, dos àmbits docents addicionals atès la seva funció integradora. En el moment de definir les directrius i organització de l’assignatura, es van establir a la Facultat de Farmàcia 27 àmbits docents. Tanmateix, les característiques del TFG quan a tipus de projectes o estudis es van establir inicialment en base a tres opcions...
Potential Accessibility of Web-based mathematical information resources
(2014-03) Centelles Velilla, Miquel; Ribera, Mireia; Rodríguez Santiago, Inmaculada
This paper presents a research concerning the conversion of non-accessible web pages containing mathematical formulae into accessible versions through an OCR (Optical Character Recognition) tool. The objective of this research is twofold. First, to establish criteria for evaluating the potential accessibility of mathematical web sites, i.e. the feasibility of converting non-accessible (non-MathML) math sites into accessible ones (Math-ML). Second, to propose a data model and a mechanism to publish evaluation results, making them available to the educational community who may use them as a quality measurement for selecting learning material. Results show that the conversion using OCR tools is not viable for math web pages mainly due to two reasons: many of these pages are designed to be interactive, making difficult, if not almost impossible, a correct conversion; formula (either images or text) have been written without taking into account standards of math writing, as a consequence OCR tools do not properly recognize math symbols and expressions. In spite of these results, we think the proposed methodology to create and publish evaluation reports may be rather useful in other accessibility assessment scenarios.

Examinar

Enviaments recents