Search

Federated learning (FL), a relatively new area of research in medical image analysis, enables collaborative learning of a federated deep learning model without sharing the data of participating clients. In this paper, we propose FedDropoutAvg, a new federated learning approach for detection of tumor in images of colon tissue slides. The proposed method leverages the power of dropout, a commonly employed scheme to avoid overfitting in neural networks, in both client selection and federated averaging processes. We examine FedDropoutAvg against other FL benchmark algorithms for two different image classification tasks using a publicly available multi-site histopathology image dataset. We train and test the proposed model on a large dataset consisting of 1.2 million image tiles from 21 different sites. For testing the generalization of all models, we select held-out test sets from sites that were not used during training. We show that the proposed approach outperforms other FL methods and reduces the performance gap (to less than 3% in terms of AUC on independent test sites) between FL and a central deep learning model that requires all data to be shared for centralized training, demonstrating the potential of the proposed FedDropoutAvg model to be more generalizable than other state-of-the-art federated models. To the best of our knowledge, ours is the first study to effectively utilize the dropout strategy in a federated setting for tumor detection in histology images., (© 2023. The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.)

Asthma, a common chronic respiratory disease among children and adults, affects more than 200 million people worldwide and causes about 450,000 deaths each year. Machine learning is increasingly applied in healthcare to assist health practitioners in decision-making. In asthma management, machine learning excels in performing well-defined tasks, such as diagnosis, prediction, medication, and management. However, there remain uncertainties about how machine learning can be applied to predict asthma exacerbation. This study aimed to systematically review recent applications of machine learning techniques in predicting the risk of asthma attacks to assist asthma control and management. A total of 860 studies were initially identified from five databases. After the screening and full-text review, 20 studies were selected for inclusion in this review. The review considered recent studies published from January 2010 to February 2023. The 20 studies used machine learning techniques to support future asthma risk prediction by using various data sources such as clinical, medical, biological, and socio-demographic data sources, as well as environmental and meteorological data. While some studies considered prediction as a category, other studies predicted the probability of exacerbation. Only a group of studies applied prediction windows. The paper proposes a conceptual model to summarise how machine learning and available data sources can be leveraged to produce effective models for the early detection of asthma attacks. The review also generated a list of data sources that other researchers may use in similar work. Furthermore, we present opportunities for further research and the limitations of the preceding studies. [ABSTRACT FROM AUTHOR]

Precise neurosurgical guidance is critical for successful brain surgeries and plays a vital role in all phases of image-guided neurosurgery (IGN). Neuronavigation software enables real-time tracking of surgical tools, ensuring their presentation with high precision in relation to a virtual patient model. Therefore, this work focuses on the development of a novel multimodal IGN system, leveraging deep learning and explainable AI to enhance brain tumor surgery outcomes. The study establishes the clinical and technical requirements of the system for brain tumor surgeries. NeuroIGN adopts a modular architecture, including brain tumor segmentation, patient registration, and explainable output prediction, and integrates open-source packages into an interactive neuronavigational display. The NeuroIGN system components underwent validation and evaluation in both laboratory and simulated operating room (OR) settings. Experimental results demonstrated its accuracy in tumor segmentation and the success of ExplainAI in increasing the trust of medical professionals in deep learning. The proposed system was successfully assembled and set up within 11 min in a pre-clinical OR setting with a tracking accuracy of 0.5 (± 0.1) mm. NeuroIGN was also evaluated as highly useful, with a high frame rate (19 FPS) and real-time ultrasound imaging capabilities. In conclusion, this paper describes not only the development of an open-source multimodal IGN system but also demonstrates the innovative application of deep learning and explainable AI algorithms in enhancing neuronavigation for brain tumor surgeries. By seamlessly integrating pre- and intra-operative patient image data with cutting-edge interventional devices, our experiments underscore the potential for deep learning models to improve the surgical treatment of brain tumors and long-term post-operative outcomes. [ABSTRACT FROM AUTHOR]

Many current medical image segmentation methods utilize convolutional neural networks (CNNs), with some extended U-Net-based networks relying on deep feature representations to achieve satisfactory results. However, due to the limited receptive fields of convolutional architectures, they are unable to explicitly model the varying range dependencies present in medical images. Recently, advancements in large kernel convolution have allowed for the extraction of a wider range of low frequency information, making this task more achievable. In this paper, we propose TBUnet for solving the problem of difficult to accurately segment lesions with heterogeneous structures and fuzzy borders, such as melanoma, colon polyps and breast cancer. The TBUnet is a pure convolutional network with three branches for extracting high frequency information, low frequency information, and boundary information, respectively. It is capable of extracting features in various areas. To fuse the feature maps from the three branches, TBUnet presents the FL (fusion layer) module, which is based on threshold and logical operation. We design the FE (feature enhancement) module on the skip-connection to emphasize the fine-grained features. In addition, our method varies the number of input channels in different branches at each stage of the network, so that the relationship between low and high frequency features can be learned. TBUnet yields 91.08 DSC on ISIC-2018 for melanoma segmentation, and achieves better performance than state-of-the-art medical image segmentation methods. Furthermore, experimental results with 82.48 DSC and 89.04 DSC obtained on the BUSI dataset and the Kvasir-SEG dataset show that TBUnet outperforms the advanced segmentation methods. Experiments demonstrate that TBUnet has excellent segmentation performance and generalisation capability. [ABSTRACT FROM AUTHOR]

Mobile health (mHealth) intervention systems can employ adaptive strategies to interact with users. Instead of designing such complex strategies manually, reinforcement learning (RL) can be used to adaptively optimize intervention strategies concerning the user's context. In this paper, we focus on the issue of overwhelming interactions when learning a good adaptive strategy for the user in RL-based mHealth intervention agents. We present a data-driven approach integrating psychological insights and knowledge of historical data. It allows RL agents to optimize the strategy of delivering context-aware notifications from empirical data when counterfactual information (user responses when receiving notifications) is missing. Our approach also considers a constraint on the frequency of notifications, which reduces the interaction burden for users. We evaluated our approach in several simulation scenarios using real large-scale running data. The results indicate that our RL agent can deliver notifications in a manner that realizes a higher behavioral impact than context-blind strategies. [ABSTRACT FROM AUTHOR]

In this paper, considering year 2020 and Covid-19, we analyze medical imaging tools and their performance scores in accordance with the dataset size and their complexity. For this, we mainly consider AI-driven tools that employ two different types of image data, namely chest Computed Tomography (CT) and X-ray. We elaborate on their strengths and weaknesses by taking the following important factors into account: i) dataset size; ii) model fitting criteria (over-fitting and under-fitting); iii) transfer learning in the deep learning era; and iv) data augmentation. Medical imaging tools do not explicitly analyze model fitting. Also, using transfer learning, with fewer data, one could possibly build Covid-19 deep learning model but they are limited to education and training. We observe that, in both image modalities, neither the dataset size nor does data augmentation work well for Covid-19 screening purposes because a large dataset does not guarantee all possible Covid-19 manifestations and data augmentation does not create new Covid-19 cases. [ABSTRACT FROM AUTHOR]

We explore the Covid-19 diffusion with an agent-based model of an Italian region with a population on a scale of 1:1000. We also simulate different vaccination strategies. From a decision support system perspective, we investigate the adoption of artificial intelligence techniques to provide suggestions about more effective policies. We adopt the widely used multi-agent programmable modeling environment NetLogo, adding genetic algorithms to evolve the best vaccination criteria. The results suggest a promising methodology for defining vaccine rates by population types over time. The results are encouraging towards a more extensive application of agent-oriented methods in public healthcare policies. [ABSTRACT FROM AUTHOR]

Obesity and overweight has increased in the last year and has become a pandemic disease, the result of sedentary lifestyles and unhealthy diets rich in sugars, refined starches, fats and calories. Machine learning (ML) has proven to be very useful in the scientific community, especially in the health sector. With the aim of providing useful tools to help nutritionists and dieticians, research focused on the development of ML and Deep Learning (DL) algorithms and models is searched in the literature. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol has been used, a very common technique applied to carry out revisions. In our proposal, 17 articles have been filtered in which ML and DL are applied in the prediction of diseases, in the delineation of treatment strategies, in the improvement of personalized nutrition and more. Despite expecting better results with the use of DL, according to the selected investigations, the traditional methods are still the most used and the yields in both cases fluctuate around positive values, conditioned by the databases (transformed in each case) to a greater extent than by the artificial intelligence paradigm used. Conclusions: An important compilation is provided for the literature in this area. ML models are time-consuming to clean data, but (like DL) they allow automatic modeling of large volumes of data which makes them superior to traditional statistics. [ABSTRACT FROM AUTHOR]

Many modifiable and non-modifiable risk factors have been associated with hypertension. However, current screening programs are still failing in identifying individuals at higher risk of hypertension. Given the major impact of high blood pressure on cardiovascular events and mortality, there is an urgent need to find new strategies to improve hypertension detection. We aimed to explore whether a machine learning (ML) algorithm can help identifying individuals predictors of hypertension. We analysed the data set generated by the questionnaires administered during the World Hypertension Day from 2015 to 2019. A total of 20206 individuals have been included for analysis. We tested five ML algorithms, exploiting different balancing techniques. Moreover, we computed the performance of the medical protocol currently adopted in the screening programs. Results show that a gain of sensitivity reflects in a loss of specificity, bringing to a scenario where there is not an algorithm and a configuration which properly outperforms against the others. However, Random Forest provides interesting performances (0.818 sensitivity – 0.629 specificity) compared with medical protocols (0.906 sensitivity – 0.230 specificity). Detection of hypertension at a population level still remains challenging and a machine learning approach could help in making screening programs more precise and cost effective, when based on accurate data collection. More studies are needed to identify new features to be acquired and to further improve the performances of ML models. [ABSTRACT FROM AUTHOR]

Petabytes of health data are collected annually across the globe in electronic health records (EHR), including significant information stored as unstructured free text. However, the lack of effective mechanisms to securely share clinical text has inhibited its full utilization. We propose a new method, DataSifterText, to generate partially synthetic clinical free-text that can be safely shared between stakeholders (e.g., clinicians, STEM researchers, engineers, analysts, and healthcare providers), limiting the re-identification risk while providing significantly better utility preservation than suppressing or generalizing sensitive tokens. The method creates partially synthetic free-text data, which inherits the joint population distribution of the original data, and disguises the location of true and obfuscated words. Under certain obfuscation levels, the resulting synthetic text was sufficiently altered with different choices, orders, and frequencies of words compared to the original records. The differences were comparable to machine-generated (fully synthetic) text reported in previous studies. We applied DataSifterText to two medical case studies. In the CDC work injury application, using privacy protection, 60.9-86.5% of the synthetic descriptions belong to the same cluster as the original descriptions, demonstrating better utility preservation than the naïve content suppressing method (45.8-85.7%). In the MIMIC III application, the generated synthetic data maintained over 80% of the original information regarding patients' overall health conditions. The reported DataSifterText statistical obfuscation results indicate that the technique provides sufficient privacy protection (low identification risk) while preserving population-level information (high utility). [ABSTRACT FROM AUTHOR]

There has been an explosive growth in research over the last decade exploring machine learning techniques for analyzing chest X-ray (CXR) images for screening cardiopulmonary abnormalities. In particular, we have observed a strong interest in screening for tuberculosis (TB). This interest has coincided with the spectacular advances in deep learning (DL) that is primarily based on convolutional neural networks (CNNs). These advances have resulted in significant research contributions in DL techniques for TB screening using CXR images. We review the research studies published over the last five years (2016-2021). We identify data collections, methodical contributions, and highlight promising methods and challenges. Further, we discuss and compare studies and identify those that offer extension beyond binary decisions for TB, such as region-of-interest localization. In total, we systematically review 54 peer-reviewed research articles and perform meta-analysis. [ABSTRACT FROM AUTHOR]

Cardiac structure contouring is a time consuming and tedious manual activity used for radiotherapeutic dose toxicity planning. We developed an automatic cardiac structure segmentation pipeline for use in low-dose non-contrast planning CT based on deep learning algorithms for small datasets. Fifty CT scans were retrospectively selected and the whole heart, ventricles and atria were contoured. A two stage deep learning pipeline was trained on 41 non contrast planning CTs, tuned with 3 CT scans and validated on 6 CT scans. In the first stage, An InceptionResNetV2 network was used to identify the slices that contained cardiac structures. The second stage consisted of three deep learning models trained on the images containing cardiac structures to segment the structures. The three deep learning models predicted the segmentations/contours on axial, coronal and sagittal images and are combined to create the final prediction. The final accuracy of the pipeline was quantified on 6 volumes by calculating the Dice similarity coefficient (DC), 95% Hausdorff distance (95% HD) and volume ratios between predicted and ground truth volumes. Median DC and 95% HD of 0.96, 0.88, 0.92, 0.80 and 0.82, and 1.86, 2.98, 2.02, 6.16 and 6.46 were achieved for the whole heart, right and left ventricle, and right and left atria respectively. The median differences in volume were -4, -1, + 5, -16 and -20% for the whole heart, right and left ventricle, and right and left atria respectively. The automatic contouring pipeline achieves good results for whole heart and ventricles. Robust automatic contouring with deep learning methods seems viable for local centers with small datasets. [ABSTRACT FROM AUTHOR]

Developers proposing new machine learning for health (ML4H) tools often pledge to match or even surpass the performance of existing tools, yet the reality is usually more complicated. Reliable deployment of ML4H to the real world is challenging as examples from diabetic retinopathy or Covid-19 screening show. We envision an integrated framework of algorithm auditing and quality control that provides a path towards the effective and reliable application of ML systems in healthcare. In this editorial, we give a summary of ongoing work towards that vision and announce a call for participation to the special issue Machine Learning for Health: Algorithm Auditing & Quality Control in this journal to advance the practice of ML4H auditing. [ABSTRACT FROM AUTHOR]

In radiology, natural language processing (NLP) allows the extraction of valuable information from radiology reports. It can be used for various downstream tasks such as quality improvement, epidemiological research, and monitoring guideline adherence. Class imbalance, variation in dataset size, variation in report complexity, and algorithm type all influence NLP performance but have not yet been systematically and interrelatedly evaluated. In this study, we investigate these factors on the performance of four types [a fully connected neural network (Dense), a long short-term memory recurrent neural network (LSTM), a convolutional neural network (CNN), and a Bidirectional Encoder Representations from Transformers (BERT)] of deep learning-based NLP. Two datasets consisting of radiologist-annotated reports of both trauma radiographs (n = 2469) and chest radiographs and computer tomography (CT) studies (n = 2255) were split into training sets (80%) and testing sets (20%). The training data was used as a source to train all four model types in 84 experiments (Fracture-data) and 45 experiments (Chest-data) with variation in size and prevalence. The performance was evaluated on sensitivity, specificity, positive predictive value, negative predictive value, area under the curve, and F score. After the NLP of radiology reports, all four model-architectures demonstrated high performance with metrics up to > 0.90. CNN, LSTM, and Dense were outperformed by the BERT algorithm because of its stable results despite variation in training size and prevalence. Awareness of variation in prevalence is warranted because it impacts sensitivity and specificity in opposite directions. [ABSTRACT FROM AUTHOR]

Endotracheal intubation (ETI) is a procedure to manage and secure an unconscious patient's airway. It is one of the most critical skills in emergency or intensive care. Regular training and practice are required for medical providers to maintain proficiency. Currently, ETI training is assessed by human supervisors who may make inconsistent assessments. This study aims at developing an automated assessment system that analyzes ETI skills and classifies a trainee into an experienced or a novice immediately after training. To make the system more available and affordable, we investigate the feasibility of utilizing only hand motion features as determining factors of ETI proficiency. To this end, we extract 18 features from hand motion in time and frequency domains, and also 12 force features for comparison. Subsequently, feature selection algorithms are applied to identify an ideal feature set for developing classification models. Experimental results show that an artificial neural network (ANN) classifier with five hand motion features selected by a correlation-based algorithm achieves the highest accuracy of 91.17% while an ANN with five force features has only 80.06%. This study corroborates that a simple assessment system based on a small number of hand motion features can be effective in assisting ETI training. [ABSTRACT FROM AUTHOR]