Showing papers by "Miguel Ángel González Ballester published in 2023"

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

Abstract: The early detection of glaucoma is essential in preventing visual impairment. Artificial intelligence (AI) can be used to analyze color fundus photographs (CFPs) in a cost-effective manner, making glaucoma screening more accessible. While AI models for glaucoma screening from CFPs have shown promising results in laboratory settings, their performance decreases significantly in real-world scenarios due to the presence of out-of-distribution and low-quality images. To address this issue, we propose the Artificial Intelligence for Robust Glaucoma Screening (AIROGS) challenge. This challenge includes a large dataset of around 113,000 images from about 60,000 patients and 500 different screening centers, and encourages the development of algorithms that are robust to ungradable and unexpected input data. We evaluated solutions from 14 teams in this paper, and found that the best teams performed similarly to a set of 20 expert ophthalmologists and optometrists. The highest-scoring team achieved an area under the receiver operating characteristic curve of 0.99 (95% CI: 0.98-0.99) for detecting ungradable images on-the-fly. Additionally, many of the algorithms showed robust performance when tested on three other publicly available datasets. These results demonstrate the feasibility of robust AI-enabled glaucoma screening.

Proximal Junction Failure in Spine Surgery: Integrating Geometrical and Biomechanical Global Descriptors Improves GAP Score-Based Assessment

Abstract: Study Design. Retrospective observational study. Objective. Biomechanical and geometrical descriptors are used to improve global alignment and proportion (GAP) prediction accuracy to detect proximal junctional failure (PJF). Summary of Background Data. PJF is probably the most important complication after sagittal imbalance surgery. The GAP score has been introduced as an effective predictor for PJF, but it fails in certain situations. In this study, 112 patient records were gathered (57 PJF; 55 controls) with biomechanical and geometrical descriptors measured to stratify control and failure cases. Patients and Methods. Biplanar EOS radiographs were used to build 3-dimensional full-spine models and determine spinopelvic sagittal parameters. The bending moment (BM) was calculated as the upper body mass times, the effective distance to the body center of mass at the adjacent upper instrumented vertebra +1. Other geometrical descriptors such as full balance index (FBI), spino-sacral angle (SSA), C7 plumb line/sacrofemoral distance ratio (C7/SFD ratio), T1-pelvic angle (TPA), and cervical inclination angle (CIA) were also evaluated. The respective abilities of the GAP, FBI, SSA, C7/SFD, TPA, CIA, body weight, body mass index, and BM to discriminate PJF cases were analyzed through receiver operating characteristic curves and corresponding areas under the curve (AUC). Results. GAP (AUC = 0.8816) and FBI (AUC = 0.8933) were able to discriminate PJF cases but the highest discrimination power (AUC = 0.9371) was achieved with BM at upper instrumented vertebra + 1. Parameter cutoff analyses provided quantitative thresholds to characterize the control and failure groups and led to improved PJF discrimination, with GAP and BM being the most important contributors. SSA (AUC = 0.2857), C7/SFD (AUC = 0.3143), TPA (AUC = 0.5714), CIA (AUC = 0.4571), body weight (AUC = 0.6319), and body mass index (AUC = 0.7716) did not adequately predict PJF. Conclusion. BM reflects the quantitative biomechanical effect of external loads and can improve GAP accuracy. Sagittal alignments and mechanical integrated scores could be used to better prognosticate the risk of PJF.

A measure of behavioral heterogeneity

Abstract: . In this paper we propose a novel way to measure behavioral heterogeneity in a population of stochastic individuals. Our measure is choice-based; it evaluates the probability that, over a randomly selected menu, the sampled choices of two sampled individuals differ. We provide axiomatic foundations for this measure and a decomposition result that separates heterogeneity into its intra- and inter-personal components.

A behavioral model of adaptation

Abstract: Adaptation refers to the process of changing behavior in response to a variation in the environment. We propose a model of an adaptive individual that contemplates two forces: on the one hand the individual benefits from adopting the ideal response to the new environment, but on the other hand, behavioral change is costly. We lay down the axiomatic foundations of the model. We then study two applications. The first studies a situation where ideal behavior depends on the response of another adaptive individual. The second analyzes the case where the ideal response is influenced by the strategic interaction in a cheap talk-like game.

Unsupervised Segmentation of Fetal Brain MRI using Deep Learning Cascaded Registration

Abstract: Accurate segmentation of fetal brain magnetic resonance images is crucial for analyzing fetal brain development and detecting potential neurodevelopmental abnormalities. Traditional deep learning-based automatic segmentation, although effective, requires extensive training data with ground-truth labels, typically produced by clinicians through a time-consuming annotation process. To overcome this challenge, we propose a novel unsupervised segmentation method based on multi-atlas segmentation, that accurately segments multiple tissues without relying on labeled data for training. Our method employs a cascaded deep learning network for 3D image registration, which computes small, incremental deformations to the moving image to align it precisely with the fixed image. This cascaded network can then be used to register multiple annotated images with the image to be segmented, and combine the propagated labels to form a refined segmentation. Our experiments demonstrate that the proposed cascaded architecture outperforms the state-of-the-art registration methods that were tested. Furthermore, the derived segmentation method achieves similar performance and inference time to nnU-Net while only using a small subset of annotated data for the multi-atlas segmentation task and none for training the network. Our pipeline for registration and multi-atlas segmentation is publicly available at https://github.com/ValBcn/CasReg.

Association of central obesity with unique cardiac remodelling in young adults born small for gestational age.

Abstract: AIMS Being born small for gestational age (SGA, 10% of all births) is associated with increased risk of cardiovascular mortality in adulthood together with lower exercise tolerance, but mechanistic pathways are unclear. Central obesity is known to worsen cardiovascular outcomes, but it is uncertain how it affects the heart in adults born SGA. We aimed to assess whether central obesity makes young adults born SGA more susceptible to cardiac remodelling and dysfunction. METHODS AND RESULTS A perinatal cohort from a tertiary university hospital in Spain of young adults (30-40 years) randomly selected, 80 born SGA (birth weight below 10th centile) and 75 with normal birth weight (controls) was recruited. We studied the associations between SGA and central obesity (measured via the hip-to-waist ratio and used as a continuous variable) and cardiac regional structure and function, assessed by cardiac magnetic resonance using statistical shape analysis. Both SGA and waist-to-hip were highly associated to cardiac shape (F = 3.94, P < 0.001; F = 5.18, P < 0.001 respectively) with a statistically significant interaction (F = 2.29, P = 0.02). While controls tend to increase left ventricular end-diastolic volumes, mass and stroke volume with increasing waist-to-hip ratio, young adults born SGA showed a unique response with inability to increase cardiac dimensions or mass resulting in reduced stroke volume and exercise capacity. CONCLUSION SGA young adults show a unique cardiac adaptation to central obesity. These results support considering SGA as a risk factor that may benefit from preventive strategies to reduce cardiometabolic risk.

Multi-Head Multi-Loss Model Calibration

Abstract: Delivering meaningful uncertainty estimates is essential for a successful deployment of machine learning models in the clinical practice. A central aspect of uncertainty quantification is the ability of a model to return predictions that are well-aligned with the actual probability of the model being correct, also known as model calibration. Although many methods have been proposed to improve calibration, no technique can match the simple, but expensive approach of training an ensemble of deep neural networks. In this paper we introduce a form of simplified ensembling that bypasses the costly training and inference of deep ensembles, yet it keeps its calibration capabilities. The idea is to replace the common linear classifier at the end of a network by a set of heads that are supervised with different loss functions to enforce diversity on their predictions. Specifically, each head is trained to minimize a weighted Cross-Entropy loss, but the weights are different among the different branches. We show that the resulting averaged predictions can achieve excellent calibration without sacrificing accuracy in two challenging datasets for histopathological and endoscopic image classification. Our experiments indicate that Multi-Head Multi-Loss classifiers are inherently well-calibrated, outperforming other recent calibration techniques and even challenging Deep Ensembles' performance. Code to reproduce our experiments can be found at \url{https://github.com/agaldran/mhml_calibration} .

Insights into Insulin Resistance and Calcification in the Myocardium in Type 2 Diabetes: A Coronary Artery Analysis

Abstract: Type 2 diabetes (T2D) is responsible for high incidence of cardiovascular (CV) complications leading to heart failure. Coronary artery region-specific metabolic and structural assessment could provide deeper insight into the extent of the disease and help prevent adverse cardiac events. Therefore, in this study, we aimed at investigating such myocardial dynamics for the first time in insulin-sensitive (mIS) and insulin-resistant (mIR) T2D patients. We targeted global and region-specific variations using insulin sensitivity (IS) and coronary artery calcifications (CACs) as CV risk factor in T2D patients. IS was computed using myocardial segmentation approaches at both baseline and after an hyperglycemic–insulinemic clamp (HEC) on [18F]FDG-PET images using the standardized uptake value (SUV) (ΔSUV = SUVHEC − SUVBASELINE) and calcifications using CT Calcium Scoring. Results suggest that some communicating pathways between response to insulin and calcification are present in the myocardium, whilst differences between coronary arteries were only observed in the mIS cohort. Risk indicators were mostly observed for mIR and highly calcified subjects, which supports previously stated findings that exhibit a distinguished exposure depending on the impairment of response to insulin, while projecting added potential complications due to arterial obstruction. Moreover, a pattern relating calcification and T2D phenotypes was observed suggesting the avoidance of insulin treatment in mIS but its endorsement in mIR subjects. The right coronary artery displayed more ΔSUV, whilst plaque was more present in the circumflex. However, differences between phenotypes, and therefore CV risk, were associated to left descending artery (LAD) translating into higher CACs regarding IR, which could explain why insulin treatment was effective for LAD at the expense of higher likelihood of plaque accumulation. Personalized approaches to assess T2D may lead to more efficient treatments and risk-prevention strategies.