AI applications to medical images: From machine learning to deep learning

PurposeArtificial intelligence (AI) models are playing an increasing role in biomedical research and healthcare services. This review focuses on challenges points to be clarified about how to develop AI applications as clinical decision support systems in the real-world context.MethodsA narrative review has been performed including a critical assessment of articles published between 1989 and 2021 that guided challenging sections.ResultsWe first illustrate the architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches. For ML/radiomics, the phases of feature selection and of training, validation, and testing are described. DL models are presented as multi-layered artificial/convolutional neural networks, allowing us to directly process images. The data curation section includes technical steps such as image labelling, image annotation (with segmentation as a crucial step in radiomics), data harmonization (enabling compensation for differences in imaging protocols that typically generate noise in non-AI imaging studies) and federated learning. Thereafter, we dedicate specific sections to: sample size calculation, considering multiple testing in AI approaches; procedures for data augmentation to work with limited and unbalanced datasets; and the interpretability of AI models (the so-called black box issue). Pros and cons for choosing ML versus DL to implement AI applications to medical imaging are finally presented in a synoptic way.ConclusionsBiomedicine and healthcare systems are one of the most important fields for AI applications and medical imaging is probably the most suitable and promising domain. Clarification of specific challenging points facilitates the development of such systems and their translation to clinical practice.     

Post Learning Sleep Improves Cognitive-Emotional Decision-Making: Evidence for a ‘Deck B Sleep Effect’ in the Iowa Gambling Task

The Iowa Gambling Task (IGT) is widely used to assess real life decision-making impairment in a wide variety of clinical populations. Our study evaluated how IGT learning occurs across two sessions, and whether a period of intervening sleep between sessions can enhance learning. Furthermore, we investigate whether pre-sleep learning is necessary for this improvement. A 200-trial version of the IGT was administered at two sessions separated by wake, sleep or sleep and wake (time-of-day control). Participants were categorized as learners and non-learners based on initial performance in session one. In session one, participants initially preferred the high-frequency reward decks B and D, however, a subset of learners decreased choice from negative expected value ‘bad’ deck B and increased choices towards with a positive expected value ‘good’ decks (decks C and D). The learners who had a period of sleep (sleep and sleep/wake control conditions) between sessions showed significantly larger reduction in choices from deck B and increase in choices from good decks compared to learners that had intervening wake. Our results are the first to show that post-learning sleep can improve performance on a complex decision-making task such as the IGT. These results provide new insights into IGT learning and have important implications for understanding the neural mechanisms of “sleeping on” a decision.     

Short sleep and late bedtimes are detrimental to educational learning and knowledge transfer: An investigation of individual differences in susceptibility

Good sleep hygiene practices, including consistent bedtimes and 7–9 h of sleep/night, are theorized to benefit educational learning. However, individuals differ in how much sleep they need, as well as in their chronotype preference. Therefore, some students may be more vulnerable to the cognitive effects of sleep loss, later bedtimes and nonpreferred times of learning than others. One prominent example is the debate regarding whether sleep loss and later bedtimes affect classroom learning more in female or male students. To inform this gender-and-sleep-loss debate, we developed a virtual college-level lecture to use in a controlled, laboratory setting. During Session 1, 78 undergraduate students were randomly assigned to take the lecture at 12:00 (noon condition) or 19:30 (evening condition). Then participants wore wristband actigraphy for 1 week to monitor average and intraindividual variability in sleep duration, bedtime and midpoint of sleep. During Session 2, participants completed a test at the same time of day as Session 1. The test included basic questions that were similar to trained concepts during the lecture (trained items) as well as integration questions that required application of learned concepts (knowledge-transfer items). Bayesian analyses supported the null hypothesis that time of learning did not affect test performance. Collapsed across time of testing, regression analyses showed that shorter sleep durations and later bedtimes explained 13% of the variance in test performance. Longer sleep durations and earlier bedtimes predicted better test performance primarily in females, younger students and morning-types. Interestingly, students with above-median fluid intelligence scores were resilient to short sleep and late bedtimes. Our findings indicate that both sleep and circadian factors should be addressed to optimize educational learning, particularly in the students who are most susceptible to sleep loss.     

Artificial intelligence and machine learning for medical imaging: A technology review

Artificial intelligence (AI) has recently become a very popular buzzword, as a consequence of disruptive technical advances and impressive experimental results, notably in the field of image analysis and processing. In medicine, specialties where images are central, like radiology, pathology or oncology, have seized the opportunity and considerable efforts in research and development have been deployed to transfer the potential of AI to clinical applications. With AI becoming a more mainstream tool for typical medical imaging analysis tasks, such as diagnosis, segmentation, or classification, the key for a safe and efficient use of clinical AI applications relies, in part, on informed practitioners. The aim of this review is to present the basic technological pillars of AI, together with the state-of-the-art machine learning methods and their application to medical imaging. In addition, we discuss the new trends and future research directions. This will help the reader to understand how AI methods are now becoming an ubiquitous tool in any medical image analysis workflow and pave the way for the clinical implementation of AI-based solutions.     

Nurturing diversity and inclusion in AI in Biomedicine through a virtual summer program for high school students

Artificial Intelligence (AI) has the power to improve our lives through a wide variety of applications, many of which fall into the healthcare space; however, a lack of diversity is contributing to limitations in how broadly AI can help people. The UCSF AI4ALL program was established in 2019 to address this issue by targeting high school students from underrepresented backgrounds in AI, giving them a chance to learn about AI with a focus on biomedicine, and promoting diversity and inclusion. In 2020, the UCSF AI4ALL three-week program was held entirely online due to the COVID-19 pandemic. Thus, students participated virtually to gain experience with AI, interact with diverse role models in AI, and learn about advancing health through AI. Specifically, they attended lectures in coding and AI, received an in-depth research experience through hands-on projects exploring COVID-19, and engaged in mentoring and personal development sessions with faculty, researchers, industry professionals, and undergraduate and graduate students, many of whom were women and from underrepresented racial and ethnic backgrounds. At the conclusion of the program, the students presented the results of their research projects at the final symposium. Comparison of pre- and post-program survey responses from students demonstrated that after the program, significantly more students were familiar with how to work with data and to evaluate and apply machine learning algorithms. There were also nominally significant increases in the students’ knowing people in AI from historically underrepresented groups, feeling confident in discussing AI, and being aware of careers in AI. We found that we were able to engage young students in AI via our online training program and nurture greater diversity in AI. This work can guide AI training programs aspiring to engage and educate students entirely online, and motivate people in AI to strive towards increasing diversity and inclusion in this field.     

Intelligent scheduling of FMSs with inductive learning capability using neural networks

With the growing uncertainty and complexity in the manufacturing environment, most scheduling problems have been proven to be NP-complete and this can degrade the performance of conventional operations research (OR) techniques. This article presents a system-attribute-oriented knowledge-based scheduling system (SAOSS) with inductive learning capability. With the rich heritage from artificial intelligence (AI), SAOSS takes a multialgorithm paradigm which makes it more intelligent, flexible, and suitable than others for tackling complicated, dynamic scheduling problems. SAOSS employs an efficient and effective inductive learning method, a continuous iterative dichotomister 3 (CID3) algorithm, to induce decision rules for scheduling by converting corresponding decision trees into hidden layers of a self-generated neural network. Connection weights between hidden units imply the scheduling heuristics, which are then formulated into scheduling rules. An FMS scheduling problem is also given for illustration. The scheduling results show that the system-attribute-oriented knowledge-based approach is capable of addressing dynamic scheduling problems.     

D1 receptor activation in the mushroom bodies rescues sleep-loss-induced learning impairments in Drosophila

BACKGROUND: Extended wakefulness disrupts acquisition of short-term memories in mammals. However, the underlying molecular mechanisms triggered by extended waking and restored by sleep are unknown. Moreover, the neuronal circuits that depend on sleep for optimal learning remain unidentified. RESULTS: Learning was evaluated with aversive phototaxic suppression. In this task, flies learn to avoid light that is paired with an aversive stimulus (quinine-humidity). We demonstrate extensive homology in sleep-deprivation-induced learning impairment between flies and humans. Both 6 hr and 12 hr of sleep deprivation are sufficient to impair learning in Canton-S (Cs) flies. Moreover, learning is impaired at the end of the normal waking day in direct correlation with time spent awake. Mechanistic studies indicate that this task requires intact mushroom bodies (MBs) and requires the dopamine D1-like receptor (dDA1). Importantly, sleep-deprivation-induced learning impairments could be rescued by targeted gene expression of the dDA1 receptor to the MBs. CONCLUSIONS: These data provide direct evidence that extended wakefulness disrupts learning in Drosophila. These results demonstrate that it is possible to prevent the effects of sleep deprivation by targeting a single neuronal structure and identify cellular and molecular targets adversely affected by extended waking in a genetically tractable model organism.     

The Consolidation of Implicit Sequence Memory in Obstructive Sleep Apnea

Obstructive Sleep Apnea (OSA) Syndrome is a relatively frequent sleep disorder characterized by disrupted sleep patterns. It is a well-established fact that sleep has beneficial effect on memory consolidation by enhancing neural plasticity. Implicit sequence learning is a prominent component of skill learning. However, the formation and consolidation of this fundamental learning mechanism remains poorly understood in OSA. In the present study we examined the consolidation of different aspects of implicit sequence learning in patients with OSA. We used the Alternating Serial Reaction Time task to measure general skill learning and sequence-specific learning. There were two sessions: a learning phase and a testing phase, separated by a 10-hour offline period with sleep. Our data showed differences in offline changes of general skill learning between the OSA and control group. The control group demonstrated offline improvement from evening to morning, while the OSA group did not. In contrast, we did not observe differences between the groups in offline changes in sequence-specific learning. Our findings suggest that disrupted sleep in OSA differently affects neural circuits involved in the consolidation of sequence learning.     

Fluctuations and synchronizations of neural activities during sleep: Neural basis of possible sleep functions?

Physiological and psychological evidence have been accumulated concerning the function of sleep in development and learning/memory. Many conceptual ideas have been proposed to elucidate the mechanisms underlying them. Sleep consists of a wide variety of physiological processes. It has not yet been clarified which processes are involved in development and learning/memory processes. We have found that single neuronal activity exhibits a slowly fluctuating rate of discharge during rapid eye movement (REM) sleep and a random low discharge rate during non-rapid eye movement (NREM) sleep. It is suggested that a structural change of the neural network attractor underlies this neuronal dynamics-alternation by mathematical modeling. Functional interpretation of the neuronal dynamics-alternation was provided in combination with the phase locking of ponto-geniculo-occipital (PGO)/pontine (P) wave to the hippocampal theta wave, each of which is known to be involved in learning/memory processes. More directly, by the long-term sensory deprivation, the dynamics of neural activity during sleep was found to progressively change in a non-monotonic way. This finding reveals a possible interaction between sleep and reorganization of neural network in the matured brain. Here, in addition to the related findings, we described our idea about how sleep contributes to the learning/memory processes and reorganization of neural network of the matured brain through characteristic neural activities during sleep.

     

基于深度学习的肝脏病理图像中肝脂肪变性分级研究

摘要 目的：设计基于深层神经网络模型用来分析肝脏全景病理切片图像（Whole slide images, WSI）的肝脂肪变性分级方法,以实现对非酒精性脂肪性肝病（Non-alcoholic fatty liver disease, NAFLD）病程的辅助诊断。方法：结合临床诊断,以非酒精性脂肪肝活动度积分（NAFLD activity score, NAS）为评价标准,将肝脂肪变性程度分为无、轻度、中度和重度等四级病程,本研究采用多示例学习的策略构建并训练深度神经网络模型,将训练获得的人工智能模型用来实现计算机自动化诊断肝脏病理切片中肝脂肪变性程度分级。结果：通过使用本研究中的人工智能方法可以在3分钟内对一张WSI进行完整的分析,得到该病患肝脏病理切片中肝脂肪变性分级,训练获得的人工智能模型的AUC为0.97,肝脂肪变性分级的平均准确率为78.18%,macro-F1 score、macro-Precision和macro-Recall分别为79.49、82.03和77.10,其结果展示获得的人工智能模型已满足可辅助临床诊断的水平。结论：本研究基于深度学习技术开发的人工智能方法初步实现快速自动化诊断肝脂肪变性分级,展现了其潜在的临床使用价值。     

Dynamics of a memory trace: effects of sleep on consolidation

BACKGROUND: There is evidence that sleep is important for memory consolidation, but the underlying neuronal changes are not well understood. We studied the effect of sleep modulation on memory and on neuronal activity in a memory system of the domestic chick brain after the learning process of imprinting. Neurons in this system become, through imprinting, selectively responsive to a training (imprinting) stimulus and so possess the properties of a memory trace. RESULTS: The proportion of neurons responsive to the training stimulus reaches a maximum the day after training. We demonstrate that sleep is necessary for this maximum to be achieved, that sleep stabilizes the initially unstable, selective responses of neurons to the imprinting stimulus, and that for sleep to be effective, it must occur during a particular period of time after training. During this period, there is a time-dependent increase in EEG activity in the 5-6 Hz band, that is, in the lower range of the theta bandwidth. The effects of sleep disturbance on consolidation cannot be attributed to fatigue or to stress. CONCLUSIONS: We establish that long-term trace consolidation requires sleep within a restricted period shortly after learning. Undisturbed sleep is necessary for the stabilization of long-term memory, measured at the behavioral and neuronal levels, and of long-term but not short-term neuronal responsiveness to the training stimulus.     

A criticality-based framework for task composition in multi-agent bioinformatics integration systems

MOTIVATION: During task composition, such as can be found in distributed query processing, workflow systems and AI planning, decisions have to be made by the system and possibly by users with respect to how a given problem should be solved. Although there is often more than one correct way of solving a given problem, these multiple solutions do not necessarily lead to the same result. Some researchers are addressing this problem by providing data provenance information. Others use expert advice encoded in a supporting knowledge-base. In this paper, we propose an approach that assesses the importance of such decisions with respect to the overall result. We present a way of measuring decision criticality and describe its potential use. RESULTS: A multi-agent bioinformatics integration system is used as the basis of a framework that facilitates such functionality. We propose an agent architecture, and a concrete bioinformatics example (prototype) is used to show how certain decisions may not be critical in the context of more complex tasks.     

Artificial intelligence in radiotherapy

Artificial intelligence (AI) has already been implemented widely in the medical field in the recent years. This paper first reviews the background of AI and radiotherapy. Then it explores the basic concepts of different AI algorithms and machine learning methods, such as neural networks, that are available to us today and how they are being implemented in radiotherapy and diagnostic processes, such as medical imaging, treatment planning, patient simulation, quality assurance and radiation dose delivery. It also explores the ongoing research on AI methods that are to be implemented in radiotherapy in the future. The review shows very promising progress and future for AI to be widely used in various areas of radiotherapy. However, basing on various concerns such as availability and security of using big data, and further work on polishing and testing AI algorithms, it is found that we may not ready to use AI primarily in radiotherapy at the moment.     

How Health Behaviors Relate to Academic Performance via Affect: An Intensive Longitudinal Study

Objective

This intensive longitudinal study examined how sleep and physical activity relate to university students’ affect and academic performance during a stressful examination period.

Methods

On 32 consecutive days, 72 first-year students answered online questionnaires on their sleep quality, physical activity, positive and negative affect, learning goal achievement, and examination grades. First-year university students are particularly well-suited to test our hypotheses: They represent a relatively homogeneous population in a natural, but controlled setting, and simultaneously deal with similar stressors, such as examinations. Data were analyzed using multilevel structural equation models.

Results

Over the examination period, better average sleep quality but not physical activity predicted better learning goal achievement. Better learning goal achievement was associated with increased probability of passing all examinations. Relations of average sleep quality and average physical activity with learning goal achievement were mediated by experienced positive affect. In terms of day-to-day dynamics, on days with better sleep quality, participants reported better learning goal achievement. Day-to-day physical activity was not related to daily learning goal achievement. Daily positive and negative affect both mediated the effect of day-to-day sleep quality and physical activity on daily learning goal achievement.

Conclusion

Health behaviors such as sleep quality and physical activity seem important for both academic performance and affect experience, an indicator of mental health, during a stressful examination period. These results are a first step toward a better understanding of between- and within-person variations in health behaviors, affect, and academic performance, and could inform prevention and intervention programs for university students.     

Importance of sleep for avian vocal communication

Sleep is one of the few truly ubiquitous animal behaviours, and though many animals spend enormous periods of time asleep, we have only begun to understand the consequences of sleep disturbances. In humans, sleep is crucial for effective communication. Birds are classic models for understanding the evolution and mechanisms of human language and speech. Bird vocalizations are remarkably diverse, critical, fitness-related behaviours, and the way sleep affects vocalizations is likely similarly varied. However, research on the effects of sleep disturbances on avian vocalizations is shockingly scarce. Consequently, there is a critical gap in our understanding of the extent to which sleep disturbances disrupt communication. Here, we argue that sleep disturbances are likely to affect all birds'' vocal performance by interfering with motivation, memory consolidation and vocal maintenance. Further, we suggest that quality sleep is likely essential when learning new vocalizations and that sleep disturbances will have especially strong effects on learned vocalizations. Finally, we advocate for future research to address gaps in our understanding of how sleep influences vocal learning and performance in birds.     

Closing the translation gap: AI applications in digital pathology

Recent advances in artificial intelligence show tremendous promise to improve the accuracy, reproducibility, and availability of medical diagnostics across a number of medical subspecialities. This is especially true in the field of digital pathology, which has recently witnessed a surge in publications describing state-of-the-art performance for machine learning models across a wide range of diagnostic applications. Nonetheless, despite this promise, there remain significant gaps in translating applications for any of these technologies into actual clinical practice. In this review, we will first give a brief overview of the recent progress in applying AI to digitized pathology images, focusing on how these tools might be applied in clinical workflows in the near term to improve the accuracy and efficiency of pathologists. Then we define and describe in detail the various factors that need to be addressed in order to successfully close the “translation gap” for AI applications in digital pathology.     

Classification des stades de sommeil par des réseaux de neurones artificiels hiérarchiques

The goal of our work is to provide an automatic analysis and decision tool for sleep stages classification based on an artificial neural networks (ANN). The first difficulty lies in choosing the physiological signals representation and in particular the electroencephalogram (EEG). Once the representation adopted, the next step is to design the optimal neural network determined by a learning and validation process of data from a set of sleep records. We studied several configurations of conventional ANN giving results varying from 62 to 71 %, then we proposed a new hierarchical configuration, which gives a rate of 74 % correct classification for six stages. These results lead us to further explore this issue at the representation and design of ANNs to improve the performance of our tool.     

AI in 3D compound design

The success of Artificial Intelligence (AI) across a wide range of domains has fuelled significant interest in its application to designing novel compounds and screening compounds against a specific target. However, many existing AI methods either do not account for the 3D structure of the target at all or struggle to capture meaningful spatial information from the target. In this Opinion, we highlight a range of recent structure-aware approaches which utilise deep learning for compound design and virtual screening. We discuss how such methods can be better integrated into existing drug discovery pipelines by facilitating the design of compounds which conform to a specified design hypothesis and by uncovering key protein-ligand interactions which can be used to aid molecule design.     

Artificial intelligence in the experimental determination and prediction of macromolecular structures

Machine learning methods, in particular convolutional neural networks, have been applied to a variety of problems in cryo-EM and macromolecular crystallographic structure solution. However, they still have only limited acceptance by the community, mainly in areas where they replace repetitive work and allow for easy visual checking, such as particle picking, crystal centering or crystal recognition. With Artificial Intelligence (AI) based protein fold prediction currently revolutionizing the field, it is clear that their scope could be much wider. However, whether we will be able to exploit this potential fully will depend on the manner in which we use machine learning: training data must be well-formulated, methods need to utilize appropriate architectures, and outputs must be critically assessed, which may even require explaining AI decisions.     

Interpretation of diagnostic data: 6. How to do it with more complex maths.

We have now shown you how to use decision analysis in making those rare, tough diagnostic decisions that are not soluble through other, easier routes. In summary, to "use more complex maths" the following steps will be useful: Create a decision tree or map of all the pertinent courses of action and their consequences. Assign probabilities to the branches of each chance node. Assign utilities to each of the potential outcomes shown on the decision tree. Combine the probabilities and utilities for each node on the decision tree. Pick the decision that leads to the highest expected utility. Test your decision for its sensitivity to clinically sensible changes in probabilities and utilities. That concludes this series of clinical epidemiology rounds. You''ve come a long way from "doing it with pictures" and are now able to extract most of the diagnostic information that can be provided from signs, symptoms and laboratory investigations. We would appreciate learning whether you have found this series useful and how we can do a better job of presenting these and other elements of "the science of the art of medicine".