LPI - Artículos de Revista

Simplifying YOLOv5 for deployment in a real crop monitoring setting

2025-09-23T06:57:07Z

Deep learning-based object detection models have become a preferred choice for crop detection tasks in crop monitoring activities due to their high accuracy and generalization capabilities. However, their high computational demand and large memory footprint pose a challenge for use on mobile embedded devices deployed in crop monitoring settings. Vari- ous approaches have been taken to minimize the computational cost and reduce the size of object detection models such as channel and layer pruning, detection head searching, back- bone optimization, etc. In this work, we approached computational lightening, model com- pression, and speed improvement by discarding one or more of the three detection scales of the YOLOv5 object detection model. Thus, we derived up to five separate fast and light models, each with only one or two detection scales. To evaluate the new models for a real crop monitoring use case, the models were deployed on NVIDIA Jetson nano and NVIDIA Jetson Orin devices. The new models achieved up to 21.4% reduction in giga floating-point operations per second (GFLOPS), 31.9% reduction in number of parameters, 30.8% reduc- tion in model size, 28.1% increase in inference speed, with only a small average accuracy drop of 3.6%. These new models are suitable for crop detection tasks since the crops are usually of similar sizes due to the high likelihood of being in the same growth stage, thus, making it sufficient to detect the crops with just one or two detection scales.

Spherical means-based free-water volume fraction from diffusion MRI increases non-linearly with age in the white matter of the healthy human brain

2026-03-26T09:24:18Z

The term free-water volume fraction (FWVF) refers to the signal fraction that could be found as the cerebrospinal fluid of the brain, which has been demonstrated as a sensitive measure that correlates with cognitive performance and various neuropathological processes. It can be quantified by properly fitting the isotropic component of the magnetic resonance (MR) signal in diffusion-sensitized sequences. Using healthy subjects (178F/109M) aged 25-94, this study examines in detail the evolution of the FWVF obtained with the spherical means technique from multi-shell acquisitions in the human brain white matter across the adult lifespan, which has been previously reported to exhibit a positive trend when estimated from single-shell data using the bi-tensor signal representation. We found evidence of a noticeably non-linear gain after the sixth decade of life, with a region-specific variate and varying change rate of the spherical means-based multi-shell FWVF parameter with age, at the same time, a heteroskedastic pattern across the adult lifespan is suggested. On the other hand, the FW corrected diffusion tensor imaging (DTI) leads to a region-dependent flattened age-related evolution of the mean diffusivity (MD) and fractional anisotropy (FA), along with a considerable reduction in their variability, as compared to the studies conducted over the standard (single-component) DTI. This way, our study provides a new perspective on the trajectory-based assessment of the brain and explains the conceivable reason for the variations observed in FA and MD parameters across the lifespan with previous studies under the standard diffusion tensor imaging.

Insight into ADHD diagnosis with deep learning on Actimetry: Quantitative interpretation of occlusion maps in age and gender subgroups

2025-02-20T13:52:28Z

Attention Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental disorder in childhood that often persists into adulthood. Objectively diagnosing ADHD can be challenging due to the reliance on subjective questionnaires in clinical assessment. Fortunately, recent advancements in artificial intelligence (AI) have shown promise in providing objective diagnoses through the analysis of medical images or activity recordings. These AI-based techniques have demonstrated accurate ADHD diagnosis; however, the growing complexity of deep learning models has introduced a lack of interpretability. These models often function as black boxes, unable to offer meaningful insights into the data patterns that characterize ADHD.

Convolution-based free-form deformation for multimodal groupwise registration

2025-02-19T13:02:32Z

Recently, an efficient implementation of convolution-based free form deformations (FFD) has been proposed for both groupwise 3D monomodal and 2D pairwise multimodal registrations. However, there is still an unmet need in the field for groupwise -D multimodal registration with L > 2. In this correspondence, we address this need and present a solution for achieving accurate registration using two popular metrics: Renyi entropy and PCA2.

Default mode network components and its relationship with anomalous self-experiences in schizophrenia: A rs-fMRI exploratory study

2025-03-07T08:55:45Z

Anomalous self-experiences (ASEs) in schizophrenia have been under research for the last 20 years. However, no neuroimage studies have provided insight of the possible biological underpinning of ASEs. In this novel approach, the connectivity within the default mode network, calculated through a ROI-based analysis of functional magnetic resonance imaging data, was correlated to the ASEs scores assessed by the Inventory of Psychotic-Like Anomalous Self-Experiences (IPASE) in a sample of 22 schizophrenia patients. The Pearson's correlation coefficients between IPASE scores and intrahemispheric connectivity of the parahippocampal gyrus with the isthmus cingulate cortex in both hemispheres, and right parahippocampal gyrus with the right rostral anterior cingulate cortex were positive and significant suggesting a relation between hyperactive functional connectivity and anomalous self-experiences intensity. Prior literature reported these areas to have a role in self-processing and consciousness as well as being anatomically connected. Further research with larger sample size and comparison with controls are needed to confirm the relationship of this connectivity with anomalous self-experiences.

Diffusion sampling schemes: A generalized methodology with nongeometric criteria

2026-03-26T09:22:27Z

Purpose:The aim of this paper is to show that geometrical criteria for designingmultishellq-space sampling procedures do not necessarily translate into recon-struction matrices with high figures of merit commonly used in the compressedsensing theory. In addition, we show that a well-known method for visitingk-space in radial three-dimensional acquisitions, namely, the Spiral Phyllotaxis,is a competitive initialization for the optimization of our nonconvex objectivefunction.Theory and Methods:We propose the gradient design method WISH (WeIght-ing SHells) which uses an objective function that accounts for weighted dis-tances between gradients withinM-tuples of consecutive shells, withMrangingbetween 1 and the maximum number of shellsS. All theM-tuples share thesame weight𝜔�M. The objective function is optimized for a sample of theseweights, using Spiral Phyllotaxis as initialization. State-of-the-art General Elec-trostatic Energy Minimization (GEEM) and Spherical Codes (SC) were used forcomparison. For the three methods, reconstruction matrices of the attenuationsignal using MAP-MRI were tested using figures of merit borrowed from theCompressed Sensing theory (namely, Restricted Isometry Property —RIP— andCoherence); we also tested the gradient design using a geometric criterion basedon Voronoi cells.Results:For RIP and Coherence, WISH got better results in at least one com-bination of weights, whilst the criterion based on Voronoi cells showed anunrelated pattern.Conclusion:The versatility provided by WISH is supported by better results.Optimization in the weight parameter space is likely to provide additionalimprovements. For a practical design with an intermediate number of gradients,our results recommend to carry out the methodology here used to determine theappropriate gradient table.

Structural brain changes in patients with persistent headache after COVID-19 resolution

2025-03-07T07:53:35Z

Headache is among the most frequently reported symptoms after resolution of COVID-19. We assessed structural brain changes using T1- and diffusion-weighted MRI processed data from 167 subjects: 40 patients who recovered from COVID- 19 but suffered from persistent headache without prior history of headache (COV), 41 healthy controls, 43 patients with episodic migraine and 43 patients with chronic migraine. To evaluate gray matter and white matter changes, morphometry parameters and diffusion tensor imaging-based measures were employed, respectively. COV patients showed significant lower cortical gray matter volume and cortical thickness than healthy subjects (p < 0.05, false discovery rate corrected) in the inferior frontal and the fusiform cortex. Lower fractional anisotropy and higher radial diffusivity (p < 0.05, family-wise error corrected) were observed in COV patients compared to controls, mainly in the corpus callosum and left hemisphere. COV patients showed higher cortical volume and thickness than migraine patients in the cingulate and frontal gyri, paracentral lobule and superior temporal sulcus, lower volume in subcortical regions and lower curvature in the precuneus and cuneus. Lower diffusion metric values in COV patients compared to migraine were identified prominently in the right hemisphere. COV patients present diverse changes in the white matter and gray matter structure. White matter changes seem to be associ- ated with impairment of fiber bundles. Besides, the gray matter changes and other white matter modifications such as axonal integrity loss seemed subtle and less pronounced than those detected in migraine, showing that persistent headache after COVID-19 resolution could be an intermediate state between normality and migraine.

Efficient estimation of propagator anisotropy and non‐Gaussianity in multishell diffusion MRI with micro‐structure adaptive convolution kernels and dual Fourier integral transforms

2026-03-26T09:22:57Z

Purpose:We seek to reformulate the so-called Propagator Anisotropy (PA) andNon-Gaussianity (NG), originally conceived for the Mean Apparent Propagatordiffusion MRI (MAP-MRI), to the Micro-Structure adaptive convolution ker-nels and dual Fourier Integral Transforms (MiSFIT). These measures describerelevant normalized features of the Ensemble Average Propagator (EAP).Theory and Methods:First, the indices, which are defined as the EAP’sdissimilarity from an isotropic (PA) or a Gaussian (NG) one, are analyticallyreformulated within the MiSFIT framework. Then a comparison between theresulting maps is drawn by means of a visual analysis, a quantitative assess-ment via numerical simulations, a test-retest study across the MICRA dataset (6subjects scanned five times) and, finally, a computational time evaluation.Results:Findings illustrate the visual similarity between the indices computedwith either technique. Evaluation against synthetic ground truth data, however,demonstrates MiSFIT’s improved accuracy. In addition, the test–retest studyreveals MiSFIT’s higher degree of reliability in most of white matter regions.Finally, the computational time evaluation shows MiSFIT’s time reduction upto two orders of magnitude.Conclusions:Despite being a direct development on the MAP-MRI represen-tation, the PA and the NG can be reliably and efficiently computed withinMiSFIT’s framework. This, together with the previous findings in the originalMiSFIT’s article, could mean the difference that definitely qualifies diffusionMRI to be incorporated into regular clinical settings.

Efficient convolution-based pairwise elastic image registration on three multimodal similarity metrics

2025-02-19T13:02:51Z

This paper proposes a complete convolutional formulation for 2D multimodal pairwise image registration problems based on free-form deformations. We have reformulated in terms of discrete 1D convolutions the evaluation of spatial transformations, the regularization term, and their gradients for three different multimodal registration metrics, namely, normalized cross correlation, mutual information, and normalized mutual information. A sufficient condition on the metric gradient is provided for further extension to other metrics. The proposed approach has been tested, as a proof of concept, on contrast-enhanced first-pass perfusion cardiac magnetic resonance images. Execution times have been compared with the corresponding execution times of the classical tensor product formulation, both on CPU and GPU. The speed-up achieved by using convolutions instead of tensor products depends on the image size and the number of control points considered, the larger those magnitudes, the greater the execution time reduction. Furthermore, the speed-up will be more significant when gradient operations constitute the major bottleneck in the optimization process.