The performance of deep-learning-based segmentation models is strongly dependent on the quality of the input data, which is frequently heterogeneous or degraded in real-world medical imaging scenarios. This work presents a data-centric algorithmic pipeline designed to improve cardiac MRI segmentation accuracy through systematic image enhancement and automatic slice-quality filtering. The proposed method is formalized as deterministic algorithm that combines image processing and supervised learning components. The approach integrates a contrast- and structure-preserving enhancement stage, based on bilateral filtering and adaptive histogram equalization, with a quality-aware selection algorithm. Slice quality is assessed using anatomical attributes extracted via YOLOv11s-based localization and a supervised classification model trained to identify diagnostically reliable images. When applied to transformer-based segmentation architectures such as ViTUNeT, the pipeline yields consistent improvements across all evaluation metrics without increasing model complexity or training cost. These findings emphasize the importance of algorithmic data curation as an effective strategy for enhancing robustness and stability in deep-learning segmentation pipelines and demonstrate the broader applicability of the proposed approach to computer-vision tasks involving heterogeneous or low-quality image datasets.

Precisely classifying cells in histological images is critical for early cancer diagnosis and tumor assessment. Traditional manual methods are time-consuming and labor-intensive for histopathologists, driving the development of automated approaches using machine learning (ML) and deep learning (DL). Convolutional neural networks (CNNs) and, more recently, vision transformers (ViTs) have demonstrated significant potential in addressing the challenges of cell classification by leveraging their ability to automatically extract and learn complex features from histological images. In this work, we evaluate multiple classification architectures applied to stained histological images to determine their effectiveness in identifying cancerous cells. We compare traditional ML models, which rely on manually extracted features such as shape and texture, against two DL-based classifiers: a CNN-based model (ResNet50) and a ViT-based model. To optimize ML models, we apply principal component analysis (PCA) to refine feature selection. Meanwhile, DL models are trained on cropped cell images using two preprocessing strategies: one that includes additional surrounding cellular context and another that uses only the cell pixels. Additionally, we investigate class balancing strategies, including downsampling and oversampling through data augmentation, to mitigate the effects of dataset imbalance. Experimental results highlight the clear advantage of DL models over traditional ML approaches. ResNet50 consistently delivers robust and reliable performance across different preprocessing strategies, confirming its effectiveness for histopathological classification tasks. Meanwhile, ViTs achieve results that are comparable to those of CNNs while demonstrating a distinct advantage in classifying underrepresented nucleus classes, likely due to their ability to capture long-range dependencies. Furthermore, incorporating the surrounding cellular environment significantly improves classification accuracy, underscoring the importance of contextual information in distinguishing between different types of nuclei.

AI inference is conflicted between Performance, developer Productivity, and device Portability–the P3 problem. Machine learning compilers (MLCs) aim to address this, but their ecosystem is fragmented, with tools that each prioritize a different issue. This paper evaluates the deployment trade-offs of PyTorch-based LLMs on NVIDIA GPUs using four intertwined prominent MLC tools: torch.compile, TensorRT, XLA, and ONNX Runtime. A dual methodology is used, leveraging synthetic PyTorch models to isolate optimizations and end-to-end benchmarks with State-of-the-Art (SOTA) models (TinyLlama-1.1B, Llama-2-7B) to measure real-world performance. Findings reveal that the peak performance of Ahead-Of-Time (AOT) compilation requires architecture-specific tools such as TensorRT-LLM, which are necessary for SOTA LLMs but are unusable for PyTorch models. As for Just-In-Time (JIT) solutions such as torch.compile and its backends, they are flexible and portable, compatible with all tested models, but they do not consistently accelerate LLMs; therefore, the choice of MLC depends on P3 considerations and model architecture.

While most patients with degenerative rotator cuff tears respond to conservative treatment, a minority progress to surgery. To anticipate these cases under class imbalance, we propose a sensitivity-constrained evolutionary feature selection framework prioritizing surgical-class recall, benchmarked against traditional methods. Two variants are proposed: (i) a single-objective search maximizing balanced accuracy and (ii) a multi-objective search also minimizing the number of selected features. Both enforce a minimum-sensitivity constraint on the minority class to limit false negatives. The dataset includes 347 patients (66 surgical, 19%) described by 28 clinical, imaging, symptom, and functional variables. We compare against 62 widely adopted pipelines, including oversampling, undersampling, hybrid resampling, cost-sensitive classifiers, and imbalance-aware ensembles. The main metric is balanced accuracy, with surgical-class F1-score as secondary. Pairwise Wilcoxon tests with a win-loss ranking assessed statistical significance. Evolutionary models rank among the top; the multi-objective variant with a Balanced Bagging Classifier performs best, achieving a mean balanced accuracy of 0.741. Selected subsets recurrently include age, tear location/severity, comorbidities, and pain/functional scores, matching clinical expectations. The constraint preserved minority-class recall without discarding or synthesizing data. Sensitivity-constrained evolutionary feature selection thus offers a data-preserving, interpretable solution for pre-surgical decision support, improving balanced performance and supporting safer triage decisions.

Background: Accurate segmentation of the left ventricular myocardium in cardiac MRI is essential for developing reliable deep learning models to diagnose left ventricular non-compaction cardiomyopathy (LVNC). This work focuses on improving the segmentation database used to train these models, enhancing the quality of myocardial segmentation for more precise model training. Methods: We present a semi-automatic framework that refines segmentations through three fundamental approaches: (1) combining neural network outputs with expert-driven corrections, (2) implementing a blob-selection method to correct segmentation errors and neural network hallucinations, and (3) employing a cross-validation process using the baseline U-Net model. Results: Applied to datasets from three hospitals, these methods demonstrate improved segmentation accuracy, with the blob-selection technique boosting the Dice coefficient for the Trabecular Zone by up to 0.06 in certain populations. Conclusions: Our approach enhances the dataset’s quality, providing a more robust foundation for future LVNC diagnostic models.

Mitosis detection is critical in histopathology for accurate diagnosis and prognosis of tumors, a topic of particular interest underscored by recent challenges. In this study, we focus on developing deep learning (DL) solutions to confront this challenge within the framework of the MIDOG challenge. Leveraging the newest MIDOG challenge dataset, the MIDOG++ dataset, we explore the efficacy of object detection models. Specifically, the RetinaNet model using fastai and PyTorch frameworks. We replicate and validate the reference work, RetinaNet, using fastai, and we propose the RetinaNet model using PyTorch. Through rigorous training and evaluation, we analyze the performance of these models in detecting mitotic figures, crucial for automating histopathological analysis and improving diagnostic accuracy. Our study demonstrates the effectiveness of the RetinaNet model in mitosis detection within histopathological images. Obtaining favorable F1 scores across the different scenarios and analyzing the relationship between different tumor types.

Left ventricular non-compaction is a cardiac condition marked by excessive trabeculae in the left ventricle's inner wall. Although various methods exist to measure these structures, the medical community still lacks consensus on the best approach. Previously, we developed DL-LVTQ, a tool based on a UNet neural network, to quantify trabeculae in this region. In this study, we expand the dataset to include new patients with Titin cardiomyopathy and healthy individuals with fewer trabeculae, requiring retraining of our models to enhance predictions. We also propose ViTUNeT, a neural network architecture combining U-Net and Vision Transformers to segment the left ventricle more accurately. Additionally, we train a YOLOv8 model to detect the ventricle and integrate it with ViTUNeT model to focus on the region of interest. Results from ViTUNet and YOLOv8 are similar to DL-LVTQ, suggesting dataset quality limits further accuracy improvements. To test this, we analyze MRI images and develop a method using two YOLOv8 models to identify and remove problematic images, leading to better results. Combining YOLOv8 with deep learning networks offers a promising approach for improving cardiac image analysis and segmentation.

Left Ventricular noncompaction (LVNC) is a recently classified form of cardiomyopathy. Although various methods have been proposed for accurately quantifying trabeculae in the left ventricle (LV), consensus on the optimal approach remains elusive. Previous research introduced DL-LVTQ, a deep learning solution for trabecular quantification based on a UNet 2D convolutional neural network (CNN) architecture and a graphical user interface (GUI) to streamline its use in clinical workflows. Building on this foundation, this work presents LVNC detector, an enhanced application designed to support cardiologists in the automated diagnosis of LVNC. The application integrates two segmentation models: DL-LVTQ and ViTUNet, the latter inspired by modern hybrid architectures combining convolutional neural networks (CNNs) and transformer-based designs. These models, implemented within an ensemble framework, leverage advancements in deep learning to improve the accuracy and robustness of magnetic resonance imaging (MRI) segmentation. Key innovations include multithreading to optimize model loading times and ensemble methods to enhance segmentation consistency across MRI slices. Additionally, the platform-independent design ensures compatibility with Windows and Linux, eliminating complex setup requirements. The LVNC detector delivers an efficient and user-friendly solution for LVNC diagnosis. It enables real-time performance and allows cardiologists to select and compare segmentation models for improved diagnostic outcomes. This work demonstrates how state-of-the-art machine learning techniques can seamlessly integrate into clinical practice to reduce human error and expedite diagnostic processes.

Accurate mitosis detection is essential for cancer diagnosis and treatment. Traditional manual counting by pathologists is time-consuming and may cause errors. This research investigates automated mitosis detection in stained histopathological images using Deep Learning (DL) techniques, particularly object detection models. We propose a two-stage object detection model based on Faster R-CNN to effectively detect mitosis within histopathological images. The stain augmentation and normalization techniques are also applied to address the significant challenge of domain shift in histopathological image analysis. The experiments are conducted using the MIDOG++ dataset, the most recent dataset from the MIDOG challenge. This research builds on our previous work, in which two one-stage frameworks, in particular on RetinaNet using fastai and PyTorch, are proposed. Our results indicate favorable F1-scores across various scenarios and tumor types, demonstrating the effectiveness of the object detection models. In addition, Faster R-CNN with stain techniques provides the most accurate and reliable mitosis detection, while RetinaNet models exhibit faster performance. Our results highlight the importance of handling domain shifts and the number of mitotic figures for robust diagnostic tools.

Large language models (LLMs) have been massively applied to many tasks, often surpassing state-of-the-art approaches. While their effectiveness in code generation has been extensively studied (e.g., AlphaCode), their potential for code detection remains unexplored. This work presents the first analysis of code detection using LLMs. Our study examines essential kernels, including matrix multiplication, convolution, fast-fourier transform and LU factorization, implemented in C/C++. We propose both a preliminary, naive prompt and a novel prompting strategy for code detection. Results reveal that conventional prompting achieves great precision but poor accuracy (67.5%, 22.5%, 79.5% and 64% for GEMM, convolution, FFT and LU factorization, respectively) due to a high number of false positives. Our novel prompting strategy substantially reduces false positives, resulting in excellent overall accuracy (91.2%, 98%, 99.7% and 99.7%, respectively). These results pose a considerable challenge to existing state-of-the-art code detection methods.

In the era of heterogeneous computing, a new paradigm called accelerator level parallelism (ALP) has emerged. In ALP, accelerators are used concurrently to provide unprecedented levels of performance and energy efficiency. To reach that there are many problems to be solved, one of the most challenging being co-execution. In this paper, we present a new scheduling framework called POAS, a general method for providing co-execution to applications. Our proposal consists of four steps: predict, optimize, adapt and schedule. With POAS, an unseen application can be executed concurrently in ALP with little effort. We evaluate POAS on a heterogeneous environment consisting of CPUs, GPUs (CUDA cores), and XPUs (Tensor cores) on two different fields, namely linear algebra (matrix multiplication benchmark) and deep learning (convolution benchmark). Our experiments prove that POAS provides excellent performance and completes the tasks within a time very close to the optimal time for the hardware and applications used, with a negligible execution time overhead. Moreover, the POAS predictor performed exceptionally well, achieving very low RMSE values for both use cases. Therefore, POAS can be a valuable tool for fully exploiting ALP and improving overall performance over offloading in heterogeneous settings.

Hyper-trabeculation or non-compaction in the left ventricle of the myocardium (LVNC) is a recently classified form of cardiomyopathy. Several methods have been proposed to quantify the trabeculae accurately in the left ventricle, but there is no general agreement in the medical community to use a particular approach. In the previous work, we proposed DL-LVTQ, a deep-learning approach for left ventricular trabecular quantification based on a U-Net CNN architecture. In this work, we have extended and adapted DL-LVTQ to cope with patients with different particularities and cardiomyopathies. Patient images were taken from different scanners and hospitals. We have modified and adapted the U-Net convolutional neural network to account for the different particularities of a heterogeneous group of patients with multiple cardiomyopathies and inherited cardiomyopathies. The inclusion of new groups of patients has increased the accuracy, specificity and Kappa values while maintaining the sensitivity of the proposed method. Therefore, a better-prepared diagnosis tool is ready for various cardiomyopathies with different characteristics. Cardiologists have considered that 98.9% of the evaluated outputs are verified clinically for diagnosis. Therefore, the high precision to segment the different cardiac structures allows us to make a robust diagnostic system objective and faster, decreasing human error and time spent.

Accurate diagnosis of Left Ventricular Noncompaction Cardiomyopathy (LVNC) is critical for proper patient treatment but remains challenging. This work improves LVNC detection by improving left ventricle segmentation in cardiac MR images. Trabeculated left ventricle indicates LVNC, but automatic segmentation is difficult. We present techniques to improve segmentation and evaluate their impact on LVNC diagnosis. Three main methods are introduced: (1) using full 800 × 800 MR images rather than 512 × 512; (2) a clustering algorithm to eliminate neural network hallucinations; (3) advanced network architectures including Attention U-Net, MSA-UNet, and U-Net++.Experiments utilize cardiac MR datasets from three different hospitals. U-Net++ achieves the best segmentation performance using 800 × 800 images, and it improves the mean segmentation Dice score by 0.02 over the baseline U-Net, the clustering algorithm improves the mean Dice score by 0.06 on the images it affected, and the U-Net++ provides an additional 0.02 mean Dice score over the baseline U-Net. For LVNC diagnosis, U-Net++ achieves 0.896 accuracy, 0.907 precision, and 0.912 F1-score outperforming the baseline U-Net. Proposed techniques enhance LVNC detection, but differences between hospitals reveal problems in improving generalization. This work provides validated methods for precise LVNC diagnosis.

Dedicated tensor accelerators demonstrate the importance of linear algebra in modern applications. Such accelerators have the potential for impressive performance gains, but require programmers to rewrite code using vendor APIs - a barrier to wider scale adoption. Recent work overcomes this by matching and replacing patterns within code, but such approaches are fragile and fail to cope with the diversity of real-world codes. We develop ATC, a compiler that uses program synthesis to map regions of code to specific APIs. The mapping space that ATC explores is combinatorially large, requiring the development of program classification, dynamic analysis, variable constraint generation and lexical distance matching techniques to make it tractable. We apply ATC to real-world tensor and linear algebra codes and evaluate them against four state-of-the-art approaches. We accelerate between 2.6x and 7x more programs, leading to over an order of magnitude performance improvement.

This work covers the PHAST Library's employment, a hardware-agnostic programming library, to a real-world application like the Caffe framework. The original implementation of Caffe consists of two different versions of the source code: one to run on CPU platforms and another one to run on the GPU side. With PHAST, we aim to develop a single-source code implementation capable of running efficiently on CPU and GPU. In this paper, we start by carrying out a basic Caffe implementation performance analysis using PHAST. Then, we detail possible performance upgrades. We find that the overall performance is dominated by few 'heavy' layers. In refining the inefficient parts of this version, we find two different approaches: improvements to the Caffe source code and improvements to the PHAST Library itself, which ultimately translates into improved performance in the PHAST version of Caffe. We demonstrate that our PHAST implementation achieves performance portability on CPUs and GPUs. With a single source, the PHAST version of Caffe provides the same or even better performance than the original version of Caffe built from two different codebases. For the MNIST database, the PHAST implementation takes an equivalent amount of time as native code in CPU and GPU. Furthermore, PHAST achieves a speedup of 51% and a 49% with the CIFAR-10 database against native code in CPU and GPU, respectively. These results provide a new horizon for software development in the upcoming heterogeneous computing era.

Different technologies and approaches exist to work around the performance portability problem. Companies and academia work together to find a way to preserve performance across heterogeneous hardware using a unified language, one language to rule them all. Intel's oneAPI appears with this idea in mind. In this article, we try the new Intel solution to approach heterogeneous programming, choosing machine learning as our case study. More precisely, we choose Caffe, a machine learning framework that was created six years ago. Nevertheless, how would it be to make Caffe again from the beginning, using a fresh new technology like oneAPI? In terms of not only the ease of programming‐because only one source code would be needed to deploy Caffe to CPUs, GPUs, FPGAs, and accelerators (platforms that oneAPI currently supports)‐but also performance, where oneAPI may be capable of taking advantage of specific hardware automatically. Is Intel's oneAPI ready to take the leap?

The race for automation has reached farms and agricultural fields. Many of these facilities use the Internet of Things technologies to automate processes and increase productivity. Besides, Machine Learning and Deep Learning allow performing continuous decision making based on data analysis. In this work, we fill a gap in the literature and present a novel architecture based on IoT and Machine Learning / Deep Learning technologies for the continuous assessment of agricultural crop quality. This architecture is divided into three layers that work together to gather, process, and analyze data from different sources to evaluate crop quality. In the experiments, the proposed approach based on data aggregation from different sources reaches a lower percentage error than considering only one source. In particular, the percentage error achieved by our approach in the test dataset was 6.59, while the percentage error achieved exclusively using data from sensors was 6.71.

This paper presents HDNN, a proof-of-concept MLIR dialect for cross-platform computing specialized in deep neural networks. As target devices, HDNN supports CPUs, GPUs and TPUs. In this paper, we provide a comprehensive description of the HDNN dialect, outlining how this novel approach aims to solve the P^3 problem of parallel programming (portability, productivity, and performance). An HDNN program is device-agnostic, i.e., only the device specifier has to be changed to run a given workload in one device or another. Moreover, HDNN has been designed to be a domain-specific language, which ultimately helps programming productivity. Finally, HDNN relies on optimized libraries for heavy, performance-critical workloads. HDNN has been evaluated against other state-of-the-art machine learning frameworks on all the hardware platforms achieving excellent performance. We conclude that the ideas and concepts used in HDNN can be crucial for designing future generation compilers and programming languages to overcome the challenges of the forthcoming heterogeneous computing era.

Heterogeneous computing is becoming mainstream in all scopes. This new era in computer architecture brings a new paradigm called Accelerator Level Parallelism (ALP). In ALP, accelerators are used concurrently to provide unprecedented levels of performance and energy efficiency. To reach that, there are many problems to be solved, one of the most challenging being co-execution. This paper develops a scheduling framework called POAS, a general method for providing co-execution to generic applications. Unlike other scheduling approaches, POAS does not directly schedule applications. Instead, it is a generic model that transforms any application to make it suitable for co-execution, so that it can be executed in ALP environments. Our proposal is composed of four differentiated steps: predict, optimize, adapt and schedule. During these phases, different modifications are implemented in the application to make it suitable to be executed in ALP environments. In this work we also apply our framework to a matrix multiplication case study, outlining the critical and most important steps to port the application with POAS. We evaluate our POAS-based implementation for matrix multiplication on a CPU/GPU/XPU environment using CPU cores, CUDA cores and tensor cores (XPU). Our experiments prove that co-execution in the studied scenario can benefit from ALP, yielding speedups of up to 45% with respect to using only one accelerator. The proven flexibility and potential of POAS make it an excellent candidate to reach ALP in future computer systems.