Juan Martinez-Alajarin

Dr. Juan Martinez-Alajarin

orcid

Research Focus & Metrics

Academic Career Length

21Years (2006-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 January 2026 in Sensors
Verified
Journal Article
Open access
PDF file
AGAlejandro Martinez Guillermo
Alejandro Martinez Guillermo
JZJuan Francisco Zapata-Perez
Juan Francisco Zapata-Perez
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
AAALICIA AREVALO-GARCIA
ALICIA AREVALO-GARCIA
This work presents the development of an Artificial Intelligence (AI)-based pipeline for patient-specific three-dimensional (3D) reconstruction from oncological magnetic resonance imaging (MRI), leveraging image-derived information to enhance the analysis process. These developments were carried out within the framework of Cella Medical Solutions, forming part of a broader initiative to improve and optimize the company’s medical-image processing pipeline. The system integrates automatic MRI sequence classification using a ResNet-based architecture and segmentation of anatomical structures with a modular nnU-Net v2 framework. The classification stage achieved over 90% accuracy and showed improved segmentation performance over prior state-of-the-art pipelines, particularly for contrast-sensitive anatomies such as the hepatic vasculature and pancreas, where dedicated vascular networks showed Dice score differences of approximately 20–22%, and for musculoskeletal structures, where the model outperformed specialized networks in several elements. In terms of computational efficiency, the complete processing of a full MRI case, including sequence classification and segmentation, required approximately four minutes on the target hardware. The integration of sequence-aware information allows for a more comprehensive understanding of MRI signals, leading to more accurate delineations than approaches without such differentiation. From a clinical perspective, the proposed method has the potential to be integrated into surgical planning workflows. The segmentation outputs were converted into a patient-specific 3D model, which was subsequently integrated into Cella’s surgical planner as a proof of concept. This process illustrates the transition from voxel-wise anatomical labels to a fully navigable 3D reconstruction, representing a step toward more robust and personalized AI-driven medical-image analysis workflows that leverage sequence-aware information for enhanced clinical utility.

29 June 2026 in Sensors
Verified
Journal Article
Open access
PDF file
JMJuan Francisco Garrido Martinez
Juan Francisco Garrido Martinez
JGJavier Garrido
Javier Garrido
JZJuan Francisco Zapata-Perez
Juan Francisco Zapata-Perez
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
Automatic 3D abdominal organ segmentation in magnetic resonance imaging presents a major challenge: data heterogeneity, caused by differences in acquisition protocol, contrast, spatial resolution, and imaging characteristics, which can degrade model generalization. This work investigates the use of FiLM (Feature-wise Linear Modulation) as a metadata-conditioning mechanism to improve the contextual adaptation of 3D segmentation networks. To this end, two reference architectures, U-Net and nnU-Net v2, were compared in both their baseline versions and their FiLM-conditioned variants. The evaluation was carried out on the CHAOS and AMOS datasets, as well as on a combined CHAOS+AMOS setting to analyze multi-domain robustness. The results show that the effect of FiLM clearly depends on the underlying architecture. In U-Net, FiLM produced favourable quantitative trends in the final post-processed predictions, with a small gain in CHAOS, a clearer improvement in AMOS, and the largest effect in the combined scenario, where the increase reached +0.0215 in clean Dice, together with reductions of −6.66 mm in HD95 and −3.85 mm in ASD. However, these improvements were not uniform across all metrics and organs. In contrast, in nnU-Net the conditioning did not provide consistent benefits and reduced performance in some scenarios, particularly in AMOS. In addition, when post-processing was applied to remove islands and disconnected components, its impact was much greater in U-Net than in nnU-Net, indicating that the latter produced cleaner and more stable segmentations directly from the raw output. Overall, this work suggests that FiLM can be useful to reinforce more generic architectures such as U-Net, especially when problem heterogeneity increases, but it was not consistently beneficial for nnU-Net, whose baseline already constituted the most robust and best-balanced model in the study. These results support the idea that the value of metadata conditioning is not universal, but instead depends on the base architecture and the generalization scenario under consideration.

1 August 2006
Verified
Conference Paper
Closed access
GDGines Domenech-Asensi
Gines Domenech-Asensi
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
RRRamon Ruiz-Merino
Ramon Ruiz-Merino
JLJose A. Lopez-Alcantud
Jose A. Lopez-Alcantud
Cardiac auscultation is, together with the electrocardiogram, the first basic analysis tool used to evaluate the functional state of the heart, as well as the first indicator used to submit the patient to a cardiologist. Nowadays there is great interest in the development of automatic systems capable of aiding the physician in the diagnosis of the heart condition by analyzing the sounds and murmurs present in the cardiac cycle. Such systems are implemented using mixed signal circuits, able to acquire the cardiac sounds and provide an objective diagnosis. In this paper, an implementation of the analog part of such a system is presented. It contains analog modules to process two input channels and provide a conditioned analog signal to the AD converter for the digital processing unit. We show the results obtained for different input signals and give estimations of power dissipations and general performance of the system

7 August 2026 in Sensors
Verified
Journal Article
Open access
PDF file
PCPablo Martínez Cegarra
Pablo Martínez Cegarra
JZJuan Francisco Zapata-Perez
Juan Francisco Zapata-Perez
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
Ischaemic stroke constitutes a leading cause of global disability. Rapid extraction of the infarct core from multimodal computed tomography perfusion (CTP) imaging guides reperfusion therapy and clinical decision-making. Deep learning algorithms automate this delineation, yet hospital translation is hindered by high-dimensional data, inter-scanner variability, and the low contrast of early ischaemia. Architectural comparisons in the literature frequently carry methodological biases originating from disparate preprocessing protocols and data partitions. This study reduces these variables by evaluating three segmentation strategies under a shared preprocessing pipeline and an identical data partition using the ISLES 2024 dataset. Three models were trained on the same 133-patient partition using a shared preprocessing pipeline based on morphological skull-stripping and modality-specific clinical intensity ranges. The data, preprocessing, and partitions are held constant across models, while framework-dependent factors (optimiser, patch size, physical field of view, spatial resampling, augmentation policy, and model capacity) remain coupled to each architecture and are therefore treated as part of the compared strategy rather than as fully isolated variables. The first of these is a single-stage 5-channel nnU-Net, followed by a two-stage cascaded nnU-Net (2 and 7 channels) and a lightweight Transformer (SegFormer3D). Evaluation on a fixed 15-patient held-out test set isolated the architectural performance. The cascade model achieved the highest Dice Similarity Coefficient (0.224). The single-stage nnU-Net provided the most precise volumetric estimation, recording an Absolute Volume Difference (AVD) of 23.70 mL and a lesion-wise F1-score of 7.60%. On the other hand, SegFormer3D returned the lowest overall metrics (DSC 0.163, AVD 27.28 mL, F1 2.30%). In the small held-out cohort, paired statistical testing did not reveal significant differences between models, so the reported orderings describe the present dataset and experimental configuration rather than a general architectural law. Within these limits, the local inductive bias of the convolutional models retained an empirical advantage over the single Transformer evaluated when processing this moderately sized neuroimaging dataset, and complex cascade topologies offered only marginal gains compared with a well-calibrated single-stage network. Although the predictive segmentation of infarcted tissue at acute stages still demands computational improvements, these results suggest that preprocessing quality is at least as decisive for clinical impact as increasing the complexity of neural architectures.

9 November 2026 in Electronics
Verified
Journal Article
Open access
PDF file
PCPablo Martínez Cegarra
Pablo Martínez Cegarra
JZJuan Francisco Zapata-Perez
Juan Francisco Zapata-Perez
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
Deploying deep learning segmentation models on embedded hardware requires reducing their numerical precision through quantization, an optimization whose aim is not to improve segmentation quality but to preserve it while reducing latency and energy consumption. Its effect, however, differs between architectures. This work studies whether quantization behaves the same in a convolutional network as in a Transformer, and how the compiler resolves the requested precision on two Jetson hardware–software configurations. Two ischemic-stroke segmentation models trained on the ISLES’24 dataset, M1 (convolutional nnU-Net, ∼30 M parameters, InstanceNorm) and M3 (SegFormer3D, a 4.5 M-parameter Transformer with 32 LayerNorm layers), are deployed on a Jetson Xavier NX (TensorRT 8.5) and a Jetson Orin Nano (TensorRT 10.3). Each model is compiled with TensorRT in FP32, FP16 and INT8 calibrated through the implicit (calibrator-based) workflow, with an additional mixed-precision variant for M3, and is evaluated on the same 15 test patients with the official ISLES’24 metrics (DSC, AVD, lesion-wise F1 and ALCD), latency, power, estimated energy per volume and memory, using paired per-patient statistics. FP16 preserves the quality of both models and provides most of the efficiency gain. In M1, 44% of the layers run in genuine INT8, and INT8 causes a small, statistically inconclusive DSC loss (0.169→0.154 on the Orin) but clearly more fragmented predictions (ALCD +6.9 lesions per patient), with almost no additional latency gain on the Orin. In M3, the calibrated INT8 engine keeps 97.7% of its layers in FP32, and the mixed-precision engine, with the LayerNorm layers forced to FP16, is turned by the compiler into a de facto FP16 engine (89.5% of its layers); within the tested implementation, these observations are consistent with the activation outliers that the literature associates with LayerNorm, but they do not establish it as the cause. The FP16 engine of M3 reproduces the reference (DSC 0.163) to the third decimal, and the FP32 engines of M1 match the original PyTorch model under the same pipeline, which validates the integrity of both conversions. Requesting INT8 does not guarantee INT8 execution: the effective precision is set by the compiler and depends on the model architecture. These results characterise the computational feasibility of the deployment, not the clinical readiness of the models.

23 May 2026 in Sensors
Verified
Journal Article
Open access
PDF file
Daniel Alcaraz OrtizDaniel Alcaraz Ortiz
Daniel Alcaraz Ortiz
JZJuan Francisco Zapata-Perez
Juan Francisco Zapata-Perez
JMJuan Martinez-Alajarin
Juan Martinez-Alajarin
Ischemic stroke remains a leading cause of global mortality and long-term neurological disability, where the “Time is Brain” paradigm dictates that rapid and accurate lesion assessment is fundamental for effective clinical intervention. While the nnU-Net v2 framework has established a new state of the art in medical image segmentation, its high computational demands and reliance on data-center-grade GPUs hinder its translation into real-time, point-of-care clinical workflows. This study presents a technical feasibility analysis of a Cloud-to-Edge optimization pipeline designed to transfer a 3D nnU-Net v2 model from a high-performance cloud environment to a resource-constrained embedded device. Experimental results showed that edge deployment was associated with a reduction in overlap-based segmentation metrics compared with the cloud reference, with Dice decreasing from approximately 0.78 to 0.67. However, TensorRT FP32 and FP16 inference produced nearly identical mean segmentation metrics, suggesting that reduced-precision inference did not introduce additional measurable degradation under the evaluated conditions. The optimized FP16 configuration achieved a processing time of 10.2 s per 3D volume, representing a 33% reduction compared with embedded FP32 inference, while operating within a low-power envelope of approximately 10–13 W. These findings support the preliminary technical feasibility of executing advanced 3D volumetric segmentation models on low-power edge hardware. Nevertheless, the evaluation was limited to an internal 25-case test subset and did not include external validation, prospective clinical assessment, or reader studies. Therefore, the proposed system should be interpreted as a preliminary deployment framework rather than a clinically validated tool for autonomous stroke imaging.