Power in small hybrid dataset in training deep-learning models: how to focus annotation effort

Deep learning (DL) model training must address the memory bottleneck to continue scaling. Processing-in-memory approaches can be a viable solution as they move computations near or into the memory, reducing substantial data movement. However, to deploy applications on such hardware, end-to-end software support is crucial for efficient computation mapping and scheduling as well as extensible code generation, but no consideration has been made for DL training workloads. In this paper, we propose XLA-NDP, a compiler and runtime solution for NDPX, a near-data processing (NDP) architecture integrated with an existing DL training framework. XLA-NDP offloads NDPX kernels and schedules them to overlap with GPU kernels to maximize parallelism based on GPU and NDPX costs, while providing a template-based code generator with low-level optimizations. The experiments showed that XLA-NDP provides up to 1.41x speedup (1.24x on average) over the GPU baseline for four DL model training.

Avoiding out-of-distribution (OOD) data is critical for training supervised machine learning models in the medical imaging domain. Furthermore, obtaining labeled medical data is difficult and expensive since it requires expert annotators like doctors, radiologists, etc. Active learning (AL) is a well-known method to mitigate labeling costs by selecting the most diverse or uncertain samples. However, current AL methods do not work well in the medical imaging domain with OOD data. We propose Diagnose (avoiDing out-of-dIstribution dAta usinG submodular iNfOrmation meaSurEs), an active learning framework that can jointly model similarity and dissimilarity, which is crucial in mining in-distribution data and avoiding OOD data at the same time. Particularly, we use a small number of data points as exemplars that represent a query set of in-distribution data points and another set of exemplars that represent a private set of OOD data points. We illustrate the generalizability of our framework by evaluating it on a wide variety of real-world OOD scenarios. Our experiments verify the superiority of Diagnose over the state-of-the-art AL methods across multiple domains of medical imaging.

Today, medical image-based processing is very much used in recent research in medical domain very effectively. So, this technical approach helps the medical expertise person to take faster decision regarding earlier diagnosis and faster decision regarding operation and next stages treatments. Level set approach is a crucial method that is used domain as computer vision. So, the normal level set approach is more error prone and creates several irregularities for the evolution in the stages of stabilities of algorithms. Here, the potential function is built and create a distance regularized term for making solution of initiating level set function in periodical way. A energy function is created by combining internal and external energy function. The main purpose of this task is medical image-based detection. The DRLSE-based segmentation approach reduces the re-initialization problems and can minimize the errors. The problems of further initialization can hamper the accuracy or rate of performance. This approach can efficiently able to initialize the level set value which is flexible to changing the time steps in a difference scheme for elimination of amount of iterations. The approach provides more than 85% overall accuracy and sensitivity scores for proper segmentation of the cardiac images in medical imaging domain. This approach can effectively segment the area of contour region to the boundary with varying number of iterations with lesser error rate.

Incremental Learning is well known machine learning approach wherein the weights of the learned model are dynamically and gradually updated to generalize on new unseen data without forgetting the existing knowledge. Incremental learning proves to be time as well as resource-efficient solution for deployment of deep learning algorithms in real world as the model can automatically and dynamically adapt to new data as and when annotated data becomes available. The development and deployment of Computer Aided Diagnosis (CAD) tools in medical domain is another scenario, where incremental learning becomes very crucial as collection and annotation of a comprehensive dataset spanning over multiple pathologies and imaging machines might take years. However, not much has so far been explored in this direction. In the current work, we propose a robust and efficient method for incremental learning in medical imaging domain. Our approach makes use of Hard Example Mining technique (which is commonly used as a solution to heavy class imbalance) to automatically select a subset of dataset to fine-tune the existing network weights such that it adapts to new data while retaining existing knowledge. We develop our approach for incremental learning of our already under test model for detecting dental caries. Further, we apply our approach to one publicly available dataset and demonstrate that our approach reaches the accuracy of training on entire dataset at once, while availing the benefits of incremental learning scenario.

BackgroundArtificial intelligence (AI) technology has the potential to transform medical practice within the medical imaging industry and materially improve productivity and patient outcomes. However, low acceptability of AI as a digital healthcare intervention among medical professionals threatens to undermine user uptake levels, hinder meaningful and optimal value-added engagement, and ultimately prevent these promising benefits from being realised. Understanding the factors underpinning AI acceptability will be vital for medical institutions to pinpoint areas of deficiency and improvement within their AI implementation strategies. This scoping review aims to survey the literature to provide a comprehensive summary of the key factors influencing AI acceptability among healthcare professionals in medical imaging domains and the different approaches which have been taken to investigate them. MethodsA systematic literature search was performed across five academic databases including Medline, Cochrane Library, Web of Science, Compendex, and Scopus from January 2013 to September 2023. This was done in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) guidelines. Overall, 31 articles were deemed appropriate for inclusion in the scoping review. ResultsThe literature has converged towards three overarching categories of factors underpinning AI acceptability including: user factors involving trust, system understanding, AI literacy, and technology receptiveness; system usage factors entailing value proposition, self-efficacy, burden, and workflow integration; and socio-organisational-cultural factors encompassing social influence, organisational readiness, ethicality, and perceived threat to professional identity. Yet, numerous studies have overlooked a meaningful subset of these factors that are integral to the use of medical AI systems such as the impact on clinical workflow practices, trust based on perceived risk and safety, and compatibility with the norms of medical professions. This is attributable to reliance on theoretical frameworks or ad-hoc approaches which do not explicitly account for healthcare-specific factors, the novelties of AI as software as a medical device (SaMD), and the nuances of human-AI interaction from the perspective of medical professionals rather than lay consumer or business end users. ConclusionThis is the first scoping review to survey the health informatics literature around the key factors influencing the acceptability of AI as a digital healthcare intervention in medical imaging contexts. The factors identified in this review suggest that existing theoretical frameworks used to study AI acceptability need to be modified to better capture the nuances of AI deployment in healthcare contexts where the user is a healthcare professional influenced by expert knowledge and disciplinary norms. Increasing AI acceptability among medical professionals will critically require designing human-centred AI systems which go beyond high algorithmic performance to consider accessibility to users with varying degrees of AI literacy, clinical workflow practices, the institutional and deployment context, and the cultural, ethical, and safety norms of healthcare professions. As investment into AI for healthcare increases, it would be valuable to conduct a systematic review and meta-analysis of the causal contribution of these factors to achieving high levels of AI acceptability among medical professionals.

Accurate diagnosis of lung nodules is essential for detection and assessment of lung cancer. The present contribution proposes a descriptive model for diagnostic classification of lung nodules by jointly using deep and spectral features from the 3D surface structure of nodules. To the best of our knowledge, this is the first work that utilizes a point cloud (PC)-based deep network for extracting nodule shape features. The PC-based deep network takes into account the 3D context of a nodule; meanwhile, it is extensively less computationally intensive. The spectral features prevent over-fitting, a common problem of deep networks trained by relatively small dataset in the medical imaging domain, and compensates for missing information of mesh connections. Experimental results reveal that our descriptive model demonstrates high sensitivity (87.23%) as well as high specificity (89.80%) with a total accuracy of 88.54% for reliable and accurate prediction of lung nodule malignancy.

MRI scans for Alzheimer's disease (AD) detection are popular. Recent computer vision (CV) and deep learning (DL) models help construct effective computer assisted diagnosis (CAD) models for AD detection and categorization. Due to their dependence on huge training datasets and effective hyper parameter tuning procedures, most models failed. Transfer learning adjusts the final fully linked layers to use trained DL models on smaller datasets. It handles picture categorization problems well. This research introduces a Wasserstein GAN-Gradient Penalty with Deep Transfer Learning (WGANGP-DTL)-based AD classification model on 3D MRI data. WGANGP-DTL aims to accurately identify and classify AD. WGANGP increases dataset size. The WGANGP-DTL model pre-processes MRI images using image enhancement and segments them using 3DSFCM. Feature extraction uses ant lion optimizer (ALO) with Inception v3 model. For classifying AD, deep belief network (DBN) model is used. The WGANGP-DTL model is experimentally validated using benchmark 3D MRI datasets. WGANGP-DTL outperformed recent methods in experiments.

Fire detection is a critical safety issue due to the major and irreversible consequences of fire, from economic prejudices to loss of life. It is therefore of utmost importance to design reliable, automated systems that can issue early alarms. The objective of this review is to present the state of the art in the area of fire detection, prevention and propagation modeling with machine learning algorithms. In order to understand how an artificial intelligence application penetrates an area of fire detection, a quantitative scientometric analysis was first performed. A literature search process was conducted on the SCOPUS database using terms and Boolean expressions related to fire detection techniques and machine learning areas. A number of 2332 documents were returned upon the bibliometric analysis. Fourteen datasets used in the training of deep learning models were examined, discussing critically the quality parameters, such as the dataset volume, class imbalance, and sample diversity. A separate discussion was dedicated to identifying issues that require further research in order to provide further insights, and faster and more accurate models.. The literature survey identified the main issues the current research should address: class imbalance in datasets, misclassification, and datasets currently used in model training. Recent advances in deep learning models such as transfer learning and (vision) transformers were discussed.

Deep learning models are notoriously data-hungry. Thus, there is an urging need for data-efficient techniques in medical image analysis, where well-annotated data are costly and time consuming to collect. Motivated by the recently revived “Copy-Paste” augmentation, we propose TumorCP, a simple but effective object-level data augmentation method tailored for tumor segmentation. TumorCP is online and stochastic, providing unlimited augmentation possibilities for tumors’ subjects, locations, appearances, as well as morphologies. Experiments on kidney tumor segmentation task demonstrate that TumorCP surpasses the strong baseline by a remarkable margin of 7.12% on tumor Dice. Moreover, together with image-level data augmentation, it beats the current state-of-the-art by 2.32% on tumor Dice. Comprehensive ablation studies are performed to validate the effectiveness of TumorCP. Meanwhile, we show that TumorCP can lead to striking improvements in extremely low-data regimes. Evaluated with only 10% labeled data, TumorCP significantly boosts tumor Dice by 21.87%. To the best of our knowledge, this is the very first work exploring and extending the “Copy-Paste” design in medical imaging domain. Code is available at: https://github.com/YaoZhang93/TumorCP.

Race against the Machine: can deep learning recognize microstructures as well as the trained human eye?

MA-SAM: Modality-agnostic SAM adaptation for 3D medical image segmentation

Toxidromes constitute patterns of symptoms and signs caused by specific toxic effects that guide emergency treatments. Computational identification of chemicals that cause different toxidromes allows us to rapidly screen novel compounds and compound classes as to their potential toxicity. The aim of the current study was to create a computational toolset that can map chemicals to their potential toxidromes. Hence, we evaluated the performance of a state-of-the-art deep learning method—the recently developed communicative message passing neural network (CMPNN)—for its ability to overcome the use of small datasets for training deep learning models. Our results indicated that multi-task training—a technique known for its ability to use multiple small datasets to train conventional deep neural networks—works equally well with CMPNN. We also showed that CMPNN-based ensemble learning results in more reliable predictions than those obtained using a single CMPNN model. In addition, we showed that the standard deviations of individual model predictions from an ensemble of CMPNN models correlated with the errors of ensemble predictions and could be used to estimate the reliability of ensemble predictions. For toxidromes that do not have well-defined molecular mechanisms or sufficient data to train a deep learning model, we used the similarity ensemble approach to develop molecular structural similarity-based toxidrome models. We made the toolset developed in this study publicly accessible via a web user interface at https://toxidrome.bhsai.org/.

DrasCLR: A self-supervised framework of learning disease-related and anatomy-specific representation for 3D lung CT images

The new challenge in Artificial Intelligence (AI) is to understand the limitations of models to reduce potential harm. Particularly, unknown disparities based on demographic factors could encrypt currently existing inequalities worsening patient care for some groups. Following PRISMA guidelines, we present a systematic review of ‘fair’ deep learning modeling techniques for natural and medical image applications which were published between year 2011 to 2021. Our search used Covidence review management software and incorporates articles from PubMed, IEEE, and ACM search engines and three reviewers independently review the manuscripts. Inter-rater agreement was 0.89 and conflicts were resolved by obtaining consensus between three reviewers. Our search initially retrieved 692 studies but after careful screening, our review included 22 manuscripts that carried four prevailing themes; ‘fair’ training dataset generation (4/22), representation learning (10/22), model disparity across institutions (5/22) and model fairness with respect to patient demographics (3/22). We benchmark the current literature regarding fairness in AI-based image analysis and highlighted the existing challenges. We observe that often discussion regarding fairness are limited to analyzing existing bias without further establishing methodologies to overcome model disparities. Based on the current research trends, exploration of adversarial learning for demographic/camera/institution agnostic models is an important direction to minimize disparity gaps for imaging. Privacy preserving approaches also present encouraging performance for both natural and medical image domain.

The Medical Segment Anything Model (MedSAM) has demonstrated strong performance in medical image segmentation, attracting increasing attention in the medical imaging domain. However, as with many prompt-based segmentation models, its performance is highly sensitive to the type and location of input prompts. This sensitivity often leads to suboptimal segmentation outcomes and necessitates labor-intensive manual prompt tuning, which hampers both efficiency and robustness. To address this challenge, this paper proposes MedSAM-U, an uncertainty-guided framework designed to automatically refine prompt inputs and enhance segmentation reliability. Specifically, a Multi-Prompt Adapter is integrated into MedSAM, resulting in MPA-MedSAM, which enables the model to effectively accommodate diverse multi-prompt inputs. An uncertainty estimation module is then introduced to evaluate the reliability of the prompts and their initial segmentation results. Based on this, a novel uncertainty-guided prompt adaptation strategy is applied to automatically generate refined prompts and more accurate segmentation outputs. The proposed MedSAM-U framework is evaluated across multiple medical imaging modalities. Experimental results on five diverse datasets demonstrate that MedSAM-U achieves consistent performance improvements ranging from 1.7% to 20.5% over the baseline MedSAM, confirming its effectiveness and practicality for robust and efficient medical image segmentation.