-
A ResNet is All You Need? Modeling A Strong Baseline for Detecting Referable Diabetic Retinopathy in Fundus Images
Authors:
Tomás Castilla,
Marcela S. Martínez,
Mercedes Leguía,
Ignacio Larrabide,
José Ignacio Orlando
Abstract:
Deep learning is currently the state-of-the-art for automated detection of referable diabetic retinopathy (DR) from color fundus photographs (CFP). While the general interest is put on improving results through methodological innovations, it is not clear how good these approaches perform compared to standard deep classification models trained with the appropriate settings. In this paper we propose…
▽ More
Deep learning is currently the state-of-the-art for automated detection of referable diabetic retinopathy (DR) from color fundus photographs (CFP). While the general interest is put on improving results through methodological innovations, it is not clear how good these approaches perform compared to standard deep classification models trained with the appropriate settings. In this paper we propose to model a strong baseline for this task based on a simple and standard ResNet-18 architecture. To this end, we built on top of prior art by training the model with a standard preprocessing strategy but using images from several public sources and an empirically calibrated data augmentation setting. To evaluate its performance, we covered multiple clinically relevant perspectives, including image and patient level DR screening, discriminating responses by input quality and DR grade, assessing model uncertainties and analyzing its results in a qualitative manner. With no other methodological innovation than a carefully designed training, our ResNet model achieved an AUC = 0.955 (0.953 - 0.956) on a combined test set of 61007 test images from different public datasets, which is in line or even better than what other more complex deep learning models reported in the literature. Similar AUC values were obtained in 480 images from two separate in-house databases specially prepared for this study, which emphasize its generalization ability. This confirms that standard networks can still be strong baselines for this task if properly trained.
△ Less
Submitted 6 October, 2022;
originally announced October 2022.
-
Assessing Coarse-to-Fine Deep Learning Models for Optic Disc and Cup Segmentation in Fundus Images
Authors:
Eugenia Moris,
Nicolás Dazeo,
Maria Paula Albina de Rueda,
Francisco Filizzola,
Nicolás Iannuzzo,
Danila Nejamkin,
Kevin Wignall,
Mercedes Leguía,
Ignacio Larrabide,
José Ignacio Orlando
Abstract:
Automated optic disc (OD) and optic cup (OC) segmentation in fundus images is relevant to efficiently measure the vertical cup-to-disc ratio (vCDR), a biomarker commonly used in ophthalmology to determine the degree of glaucomatous optic neuropathy. In general this is solved using coarse-to-fine deep learning algorithms in which a first stage approximates the OD and a second one uses a crop of thi…
▽ More
Automated optic disc (OD) and optic cup (OC) segmentation in fundus images is relevant to efficiently measure the vertical cup-to-disc ratio (vCDR), a biomarker commonly used in ophthalmology to determine the degree of glaucomatous optic neuropathy. In general this is solved using coarse-to-fine deep learning algorithms in which a first stage approximates the OD and a second one uses a crop of this area to predict OD/OC masks. While this approach is widely applied in the literature, there are no studies analyzing its real contribution to the results. In this paper we present a comprehensive analysis of different coarse-to-fine designs for OD/OC segmentation using 5 public databases, both from a standard segmentation perspective and for estimating the vCDR for glaucoma assessment. Our analysis shows that these algorithms not necessarily outperfom standard multi-class single-stage models, especially when these are learned from sufficiently large and diverse training sets. Furthermore, we noticed that the coarse stage achieves better OD segmentation results than the fine one, and that providing OD supervision to the second stage is essential to ensure accurate OC masks. Moreover, both the single-stage and two-stage models trained on a multi-dataset setting showed results in pair or even better than other state-of-the-art alternatives, while ranking first in REFUGE for OD/OC segmentation. Finally, we evaluated the models for vCDR prediction in comparison with six ophthalmologists on a subset of AIROGS images, to understand them in the context of inter-observer variability. We noticed that vCDR estimates recovered both from single-stage and coarse-to-fine models can obtain good glaucoma detection results even when they are not highly correlated with manual measurements from experts.
△ Less
Submitted 8 February, 2023; v1 submitted 28 September, 2022;
originally announced September 2022.
-
Reconstructing dynamical networks via feature ranking
Authors:
Marc G. Leguia,
Zoran Levnajic,
Ljupco Todorovski,
Bernard Zenko
Abstract:
Empirical data on real complex systems are becoming increasingly available. Parallel to this is the need for new methods of reconstructing (inferring) the topology of networks from time-resolved observations of their node-dynamics. The methods based on physical insights often rely on strong assumptions about the properties and dynamics of the scrutinized network. Here, we use the insights from mac…
▽ More
Empirical data on real complex systems are becoming increasingly available. Parallel to this is the need for new methods of reconstructing (inferring) the topology of networks from time-resolved observations of their node-dynamics. The methods based on physical insights often rely on strong assumptions about the properties and dynamics of the scrutinized network. Here, we use the insights from machine learning to design a new method of network reconstruction that essentially makes no such assumptions. Specifically, we interpret the available trajectories (data) as features, and use two independent feature ranking approaches -- Random forest and RReliefF -- to rank the importance of each node for predicting the value of each other node, which yields the reconstructed adjacency matrix. We show that our method is fairly robust to coupling strength, system size, trajectory length and noise. We also find that the reconstruction quality strongly depends on the dynamical regime.
△ Less
Submitted 26 August, 2019; v1 submitted 11 February, 2019;
originally announced February 2019.
-
Dynamics of new strain emergence on a temporal network
Authors:
Sukankana Chakraborty,
Xavier R. Hoffmann,
Marc G. Leguia,
Felix Nolet,
Elisenda Ortiz,
Ottavia Prunas,
Leonardo Zavojanni,
Eugenio Valdano,
Chiara Poletto
Abstract:
Multi-strain competition on networks is observed in many contexts, including infectious disease ecology, information dissemination or behavioral adaptation to epidemics. Despite a substantial body of research has been developed considering static, time-aggregated networks, it remains a challenge to understand the transmission of concurrent strains when links of the network are created and destroye…
▽ More
Multi-strain competition on networks is observed in many contexts, including infectious disease ecology, information dissemination or behavioral adaptation to epidemics. Despite a substantial body of research has been developed considering static, time-aggregated networks, it remains a challenge to understand the transmission of concurrent strains when links of the network are created and destroyed over time. Here we analyze how network dynamics shapes the outcome of the competition between an initially endemic strain and an emerging one, when both strains follow a susceptible-infected-susceptible dynamics, and spread at time scales comparable with the network evolution one. Using time-resolved data of close-proximity interactions between patients admitted to a hospital and medical health care workers, we analyze the impact of temporal patterns and initial conditions on the dominance diagram and coexistence time. We find that strong variations in activity volume cause the probability that the emerging strain replaces the endemic one to be highly sensitive to the time of emergence. The temporal structure of the network shapes the dominance diagram, with significant variations in the replacement probability (for a given set of epidemiological parameters) observed from the empirical network and a randomized version of it. Our work contributes towards the description of the complex interplay between competing pathogens on temporal networks.
△ Less
Submitted 11 May, 2018;
originally announced May 2018.
