Paper-based analytical devices (PADs) employing colorimetric detection and smartphone images have gained wider acceptance in a variety of measurement applications. PADs are primarily meant to be used field settings where assay imaging conditions greatly vary, resulting less accurate results. Recently, machine-learning (ML)-assisted models been image analysis. We evaluated combination four ML models-logistic regression, support vector machine (SVM), random forest, artificial neural network...

Label noise is inevitable in medical image databases developed for deep learning due to the inter-observer variability caused by different levels of expertise experts annotating images, and, some cases, automated methods that generate labels from reports. It known incorrect annotations or label can degrade actual performance supervised models and bias model's evaluation. Existing literature show one class has minimal impact on another natural classification problems where target classes have...

Medical Vision Language Pretraining (VLP) has recently emerged as a promising solution to the scarcity of labeled data in medical domain. By leveraging paired/unpaired vision and text datasets through self-supervised learning, models can be trained acquire vast knowledge learn robust feature representations. Such pretrained have potential enhance multiple downstream tasks simultaneously, reducing dependency on data. However, despite recent progress its potential, there is no such...

Label noise in medical image classification datasets significantly hampers the training of supervised deep learning methods, undermining their generalizability. The test performance a model tends to decrease as label rate increases. Over recent years, several methods have been proposed mitigate impact and enhance robustness model. Predominantly, these works employed CNN-based architectures backbone classifiers for feature extraction. However, Vision Transformer (ViT)-based backbones replaced...

Accurate left atrium (LA) segmentation from pre-operative scans is crucial for diagnosing atrial fibrillation, treatment planning, and supporting surgical interventions. While deep learning models are key in medical image segmentation, they often require extensive manually annotated data. Foundation trained on larger datasets have reduced this dependency, enhancing generalizability robustness through transfer learning. We explore DINOv2, a self-supervised vision transformer natural images,...

Noisy labels can significantly impact medical image classification, particularly in deep learning, by corrupting learned features. Self-supervised pretraining, which doesn't rely on labeled data, enhance robustness against noisy labels. However, this varies based factors like the number of classes, dataset complexity, and training size. In images, subtle inter-class differences modality-specific characteristics add complexity. Previous research hasn't comprehensively explored interplay...

Label noise in medical image classification datasets significantly hampers the training of supervised deep learning methods, undermining their generalizability. The test performance a model tends to decrease as label rate increases. Over recent years, several methods have been proposed mitigate impact and enhance robustness model. Predominantly, these works employed CNN-based architectures backbone classifiers for feature extraction. However, Vision Transformer (ViT)-based backbones replaced...

The robustness of supervised deep learning-based medical image classification is significantly undermined by label noise. Although several methods have been proposed to enhance performance in the presence noisy labels, they face some challenges: 1) a struggle with class-imbalanced datasets, leading frequent overlooking minority classes as samples; 2) singular focus on maximizing using without incorporating experts-in-the-loop for actively cleaning labels. To mitigate these challenges, we...

Mental States are a function of brain activity; with advancements in Brain Computer Interface (BCI) tools, they can be effectively predicted. Generally, BCI researches sophisticated requiring multi-channel electrodes, and often carried out controlled lab environment. This paper illustrates that simple research, targeting specific region brain, conducted an elementary setup. A method is demonstrated to predict the level attentiveness using Electroencephalography (EEG) signal obtained from...

Correct evaluation and treatment of Scoliosis require accurate estimation spinal curvature. Current gold standard is to manually estimate Cobb Angles in X-ray images which time consuming has high inter-rater variability. We propose an automatic method with a novel framework that first detects vertebrae as objects followed by landmark detector estimates the 4 corners each vertebra separately. are calculated using slope obtained from predicted landmarks. For inference on test data, we perform...

Acquiring properly annotated data is expensive in the medical field as it requires experts, time-consuming protocols, and rigorous validation. Active learning attempts to minimize need for large samples by actively sampling most informative examples annotation. These contribute significantly improving performance of supervised machine models, thus, active can play an essential role selecting appropriate information deep learning-based diagnosis, clinical assessments, treatment planning....

Noisy labels hurt deep learning-based supervised image classification performance as the models may overfit noise and learn corrupted feature extractors. For natural training with noisy labeled data, model initialization contrastive self-supervised pretrained weights has shown to reduce corruption improve performance. However, no works have explored: i) how other approaches, such pretext task-based pretraining, impact learning label, ii) any pretraining methods alone for medical images in...

Multi-domain image-to-image translation with conditional Generative Adversarial Networks (GANs) can generate highly photo realistic images desired target classes, yet these synthetic have not always been helpful to improve downstream supervised tasks such as image classification. Improving examples requires generating high fidelity the unknown distribution of class, which many labeled GANs attempt achieve by adding soft-max cross-entropy loss based auxiliary classifier in discriminator. As...

Multi-domain image-to-image translation with conditional Generative Adversarial Networks (GANs) can generate highly photo realistic images desired target classes, yet these synthetic have not always been helpful to improve downstream supervised tasks such as image classification. Improving examples requires generating high fidelity the unknown distribution of class, which many labeled GANs attempt achieve by adding soft-max cross-entropy loss based auxiliary classifier in discriminator. As...

Paper-based analytical devices (PADs) employing colorimetric detection and smartphone images have gained wider acceptance in a variety of measurement applications. The PADs are primarily meant to be used field settings where assay imaging conditions greatly vary resulting less accurate results. Recently, machine learning (ML) assisted models been image analysis. We evaluated combinations four ML - logistic regression, support vector machine, random forest, artificial neural network, three...

Paper-based analytical devices (PADs) employing colorimetric detection and smartphone images have gained wider acceptance in a variety of measurement applications. The PADs are primarily meant to be used field settings where assay imaging conditions greatly vary resulting less accurate results. Recently, machine learning (ML) assisted models been image analysis. We evaluated combinations four ML - logistic regression, support vector machine, random forest, artificial neural network, three...