Human Experts Research Articles

Utilizing ChatGPT in Systematic Reviews and Meta-Analyses

The current uses, potential risks, and practical recommendations for using chat generative pre-trained transformers (ChatGPT) in systematic reviews (SRs) and meta-analyses (MAs) are reviewed in this article. The findings of prior research suggest that, for tasks such as literature screening and information extraction, ChatGPT can match or exceed the performance of human experts. However, for complex tasks such as risk of bias assessment, its performance remains significantly limited, underscoring the critical role of human expertise. The use of ChatGPT as an adjunct tool in SRs and MAs requires careful planning and the implementation of strict quality control and validation mechanisms to mitigate potential errors such as those arising from artificial intelligence (AI) 'hallucinations'. This paper also provides specific recommendations for optimizing human-AI collaboration in SRs and MAs. Assessing the specific context of each task and implementing the most appropriate strategies are critical when using ChatGPT in support of research goals. Furthermore, transparency regarding the use of ChatGPT in research reports is essential to maintaining research integrity. Close attention to ethical norms, including issues of privacy, bias, and fairness, is also imperative. Finally, from a human-centered perspective, this paper emphasizes the importance of researchers cultivating continuous self-iteration, prompt engineering skills, critical thinking, cross-disciplinary collaboration, and ethical awareness skills with the goals of: continuously optimizing human-AI collaboration models within reasonable and compliant norms, enhancing the complex-task performance of AI tools such as ChatGPT, and, ultimately, achieving greater efficiency through technological innovative while upholding scientific rigor.

Evaluating ASD in children through automatic analysis of paintings

Autism spectrum disorder (ASD) is a hereditary neurodevelopmental disorder affecting individuals, families, and societies worldwide. Screening for ASD relies on specialized medical resources, and current machine learning-based screening methods depend on expensive professional devices and algorithms. Therefore, there is a critical need to develop accessible and easily implementable methods for ASD assessment. In this study, we are committed to finding such an ASD screening and rehabilitation assessment solution based on children’s drawings. From an ASD painting database, 375 drawings from children with ASD and 160 drawings from typically developing children were selected, and a series of image signal processing algorithms based on typical characteristics of children with ASD were designed to extract features from images. The effectiveness of extracted features was evaluated through statistical methods, and they were then classified using a support vector machine (SVM) and XGBoost (eXtreme Gradient Boosting). In 5-fold cross-validation, the SVM achieved a recall of 94.93%, a precision of 86.40%, an accuracy of 85.98%, and an AUC of 90.90%, while the XGBoost achieved a recall of 96.27%, a precision of 93.78%, an accuracy of 92.90%, and an AUC of 98.00%. This efficacy persists at a high level even during additional validation on a set of newly collected paintings. Not only did the performance surpass that of participated human experts, but the high recall rate, as well as its affordability, manageability, and ease of implementation, indicates potentiality in wide screening and rehabilitation assessment. All analysis code is public at GitHub: dishangti/ASD-Painting-Pub.

Artificial intelligence for automatic and objective assessment of competencies in flexible bronchoscopy.

Bronchoscopy is a challenging technical procedure, and assessment of competence currently relies on expert raters. Human rating is time consuming and prone to rater bias. The aim of this study was to evaluate if a bronchial segment identification system based on artificial intelligence (AI) could automatically, instantly, and objectively assess competencies in flexible bronchoscopy in a valid way. Participants were recruited at the Clinical Skills Zone of the European Respiratory Society Annual Conference in Milan, 9th-13th September 2023. The participants performed one full diagnostic bronchoscopy in a simulated setting and were rated immediately by the AI according to its four outcome measures: diagnostic completeness (DC), structured progress (SP), procedure time (PT), and mean intersegmental time (MIT). The procedures were video-recorded and rated after the conference by two blinded, expert raters using a previously validated assessment tool with nine items regarding anatomy and dexterity. Fifty-two participants from six different continents were included. All four outcome measures of the AI correlated significantly with the experts' anatomy-ratings (Pearson's correlation coefficient, P value): DC (r=0.47, P<0.001), SP (r=0.57, P<0.001), PT (r=-0.32, P=0.02), and MIT (r=-0.55, P<0.001) and also with the experts' dexterity-ratings: DC (r=0.38, P=0.006), SP (r=0.53, P<0.001), PT (r=-0.34, P=0.014), and MIT (r=-0.47, P<0.001). The study provides initial validity evidence for AI-based immediate and automatic assessment of anatomical and navigational competencies in flexible bronchoscopy. SP provided stronger correlations with human experts' ratings than the traditional DC.

AI Telco Research: Advancements in Telecommunications Scientific Discovery

This article explores the transformative impact of Artificial Intelligence (AI) on telecommunications research and development. It examines how the integration of AI technologies is revolutionizing various aspects of the industry, including network optimization, spectrum management, and scientific discovery. The article highlights the synergy between human expertise and AI capabilities in enhancing research processes, improving data analysis, and driving innovation. Key areas of focus include AI-powered data analysis, network behavior simulation, signal processing enhancements, and the evolving nature of human-AI collaboration in telco research. The article also addresses challenges such as data privacy and algorithmic bias, while emphasizing the potential of AI to unlock new possibilities in telecommunications, from self-optimizing networks to quantum communication systems.

Effects of artificial intelligence on financial reporting accuracy

Intelligence (AI) has emerged as a transformative force in modern financial reporting, promising to revolutionize accuracy and efficiency across various industries. This study delves into the effects of AI on financial reporting accuracy, addressing critical questions surrounding its implementation, challenges, and best practices. Through a comprehensive investigation, the research aims to provide valuable insights to guide organizations in leveraging AI effectively while maintaining the integrity of their financial reporting practices. The main objective of this study is to explore the effects of artificial intelligence on the accuracy of financial reporting, examining both the benefits and challenges associated with its integration into organizational processes. Specific Objectives; To analyze how various AI technologies influence the accuracy of financial data and reporting in organizations, to explore the challenges and limitations faced by organizations when integrating AI into their financial reporting systems, to assess the importance of human oversight in ensuring the accuracy of AI-generated financial reports, to develop best practices for organizations to enhance the accuracy of financial reporting when using AI technologies. This study employed a mixed-methods approach, combining qualitative and quantitative data collection techniques to comprehensively explore the effects of AI on financial reporting accuracy. Quantitative data was gathered through surveys distributed to accountants, finance professionals, auditors, and personnel from manufacturing and tourism sectors across various industries. The surveys focused on assessing perceptions of AI's impact on financial reporting accuracy and included Likert-scale questions to gauge agreement levels. Qualitative data was obtained through in-depth interviews with selected participants to gain deeper insights into their experiences and perspectives regarding AI technologies in financial reporting. Thematic analysis was applied to interview transcripts to identify recurring themes and patterns related to AI's effects on accuracy. Participants were informed about the study's purpose and their rights, including confidentiality and anonymity. Informed consent was obtained prior to data collection, ensuring ethical standards were adhered to throughout the research process. Survey responses indicated a generally positive perception of AI's impact on financial reporting accuracy, with a majority of respondents acknowledging improvements in efficiency and error reduction. However, challenges such as data security concerns and the need for skilled personnel were highlighted as significant barriers to AI integration. Human oversight emerged as a crucial factor in validating AI-generated outputs, emphasizing the complementary role of human judgment alongside technological advancements. The findings underscored AI's potential to enhance financial reporting accuracy through advanced data analytics and automation. Key recommendations include investing in comprehensive training programs for staff, integrating AI with human expertise, implementing robust data governance frameworks, conducting regular audits of AI systems, and engaging stakeholders throughout the integration process. In conclusion, this study provided valuable insights into how AI technologies can improve the accuracy of financial reporting while addressing challenges and emphasizing the importance of human oversight. By adopting recommended best practices, organizations can maximize the benefits of AI in financial reporting, paving the way for more reliable and informed decision-making in the digital age. This research contributes to the growing body of knowledge on AI's impact on financial practices, offering practical recommendations for organizations aiming to leverage technology effectively in their financial reporting processes.

The hype of artificial intelligence in neonatology: How close are we to real AI?

Background. Artificial intelligence (AI) use in medicine has rapidly evolved, offering significant advancements in patient monitoring and diagnosis. However, the application of AI in neonatology presents unique challenges compared to adult medicine. While adult patients can often be monitored and treated autonomously through AI-powered tools like telemedicine, newborns, particularly preterm infants, are entirely dependent on continuous human care. In neonatology, human interaction—especially the tactile and empathetic care provided by nurses and doctors—remains critical, and AI systems are not equipped to replace this essential element of care. Concerns that AI might displace medical personnel in neonatal settings are unfounded, as human intervention remains irreplaceable. Discussion. Current AI technologies, often referred to as AI, are more accurately described as advanced machine learning algorithms and sophisticated software rather than true general artificial intelligence (AGI). These systems can efficiently perform well-defined tasks, such as monitoring vital signs and adjusting mechanical ventilators. However, they lack the broader understanding and adaptability necessary for independent clinical decision-making. They cannot interpret complex clinical contexts or address multifactorial medical decisions, which are still the domain of human expertise. Despite the limitations, AI holds significant potential in neonatology. It can assist in optimizing treatment protocols, such as adjusting mechanical ventilation in real-time or personalizing antibiotic treatments based on microbiological data. Furthermore, clinical decision support systems (CDSS) powered by AI can provide clinicians with evidence-based recommendations, thereby enhancing decision-making processes and improving patient outcomes. Conclusion. While promising, AI in neonatology remains a supplementary tool rather than a replacement for human judgment. Its role will likely expand, but direct human care and expertise will continue to be essential in neonatal medicine, mitigating concerns about job displacement among healthcare professionals.

Mathematical Modeling of the Global Engineering Process for Optimizing Product Quality in the Aerospace Industry

The aerospace industry faces the challenge of maintaining product excellence amidst intricate processes and demands for cost and time efficiency. Mathematical modeling emerges as a valuable tool for optimizing the engineering process and enhancing quality, with potential applications extending beyond aerospace to other sectors with high quality and safety standards. This study develops and validates a mathematical model specific to the aerospace industry, aiming to assess the impact of human resource expertise on product quality. Through a case study within an aerospace organization, an IDEF0-methodology-based mathematical model, coupled with weighted averages, was constructed to depict the comprehensive engineering process and quantify knowledge’s impact on deliverable quality. Simulation data, gathered through human resource knowledge assessments and non-conformity analyses, revealed a direct correlation between technical knowledge levels and deliverable quality, consequently impacting final product quality. The proposed model serves as a tool for estimating potential deliverable error rates and pinpointing critical areas within the process that necessitate refinement. The research underscores the significance of knowledge investment and effective knowledge management strategies in upholding quality and competitiveness across industries with stringent quality requirements.

Evaluating Large Language Models for Material Selection

Abstract Material selection is a crucial step in conceptual design due to its significant impact on the functionality, aesthetics, manufacturability, and sustainability impact of the final product. This study investigates the use of Large Language Models (LLMs) for material selection in the product design process and compares the performance of LLMs against expert choices for various design scenarios. By collecting a dataset of expert material preferences, the study provides a basis for evaluating how well LLMs can align with expert recommendations through prompt engineering and hyperparameter tuning. The divergence between LLM and expert recommendations is measured across different model configurations, prompt strategies, and temperature settings. This approach allows for a detailed analysis of factors influencing the LLMs' effectiveness in recommending materials. The results from this study highlight two failure modes, and identify parallel prompting as a useful prompt-engineering method when using LLMs for material selection. The findings further suggest that, while LLMs can provide valuable assistance, their recommendations often vary significantly from those of human experts. This discrepancy underscores the need for further research into how LLMs can be better tailored to replicate expert decision-making in material selection. This work contributes to the growing body of knowledge on how LLMs can be integrated into the design process, offering insights into their current limitations and potential for future improvements.

Repeatable Perovskite Solar Cells through Fully Automated Spin-Coating and Quenching.

Enhancing reproducibility, repeatability, as well as facilitating transferability between laboratories will accelerate the progress in many material domains, wherein perovskite-based optoelectronics are a prime use case. This study presents fully automated perovskite thin film processing using a commercial spin-coating robot in an inert atmosphere. We successfully apply this novel processing method to antisolvent quenching. This process is typically difficult to reproduce and transfer and is now enhanced to exceptional repeatability in comparison to manual processing. Champion perovskite solar cells demonstrate power conversion efficiencies as high as 19.9%, proving the transferability of established manual spin-coating processes to automatic setups. Comparison with human experts reveals that the performance is already on par, while automated processing yields improved homogeneity across the substrate surface. This work demonstrates that fully automated perovskite thin film processing improves repeatability. Such systems bear the potential to become a foundation for autonomous optimization and greatly improve transferability between laboratories.

The Role of Artificial Intelligence in Literary Analysis: A Computational Approach to Understand Literary Styles

This research explores the evolving landscape of literary analysis through the integration of Artificial Intelligence (AI) and traditional human scholarship. The primary objective is to assess the extent to which AI can enhance the analysis of literary texts by examining its performance in uncovering thematic and stylistic elements within William Shakespeare's "Hamlet." This study employs a mixedmethods research approach, combining qualitative and quantitative techniques to provide a comprehensive evaluation. In the digital age, AI has emerged as a promising tool for text analysis, offering efficiency and scalability. However, it raises fundamental questions about its ability to grasp the profound nuances, cultural contexts, and thematic richness inherent in literary works. Through meticulous comparative and thematic analyses, this research investigates the strengths and limitations of AI in literary analysis, juxtaposing its findings with traditional human interpretations. The results of our study reveal that AI excels in identifying patterns, themes, and stylistic markers within "Hamlet." It effectively recognizes key themes such as revenge, madness, and moral corruption. However, AI's analysis often lacks the depth and contextual understanding present in traditional critiques, particularly in interpreting abstract motifs and cultural references. Our findings underscore the complementary nature of AI and human scholarship in literary analysis. While AI offers quantitative efficiency and objectivity, human interpretation provides the depth, cultural insights, and emotional resonance necessary for a comprehensive understanding of literary works. We argue for a harmonious future where AI augments human expertise, leading to more profound insights and a richer literary scholarship. This research not only contributes to the field of literary analysis but also offers a broader perspective on the evolving relationship between technology and human creativity. As AI technologies advance, the collaborative synergy between AI's quantitative efficiency and human interpretation's qualitative depth promises to reshape the landscape of literary studies, enriching our understanding of literature across diverse genres, time periods, and cultural contexts.

1.5 million materials narratives generated by chatbots

The advent of artificial intelligence (AI) has enabled a comprehensive exploration of materials for various applications. However, AI models often prioritize frequently encountered material examples in the scientific literature, limiting the selection of suitable candidates based on inherent physical and chemical attributes. To address this imbalance, we generated a dataset consisting of 1,453,493 natural language-material narratives from OQMD, Materials Project, JARVIS, and AFLOW2 databases based on ab initio calculation results that are more evenly distributed across the periodic table. The generated text narratives were then scored by both human experts and GPT-4, based on three rubrics: technical accuracy, language and structure, and relevance and depth of content, showing similar scores but with human-scored depth of content being the most lagging. The integration of multimodal data sources and large language models holds immense potential for AI frameworks to aid the exploration and discovery of solid-state materials for specific applications of interest.

Redefining Spectral Data Analysis with Immersive Analytics: Exploring Domain-Shifted Model Spaces for Optimal Model Selection.

Modern developments in autonomous chemometric machine learning technology strive to relinquish the need for human intervention. However, such algorithms developed and used in chemometric multivariate calibration and classification applications exclude crucial expert insight when difficult and safety-critical analysis situations arise, e.g., spectral-based medical decisions such as noninvasively determining if a biopsy is cancerous. The prediction accuracy and interpolation capabilities of autonomous methods for new samples depend on the quality and scope of their training (calibration) data. Specifically, analysis patterns within target data not captured by the training data will produce undesirable outcomes. Alternatively, using an immersive analytic approach allows insertion of human expert judgment at key machine learning algorithm junctures forming a sensemaking process performed in cooperation with a computer. The capacity of immersive virtual reality (IVR) environments to render human comprehensible three-dimensional space simulating real-world encounters, suggests its suitability as a hybrid immersive human-computer interface for data analysis tasks. Using IVR maximizes human senses to capitalize on our instinctual perception of the physical environment, thereby leveraging our innate ability to recognize patterns and visualize thresholds crucial to reducing erroneous outcomes. In this first use of IVR as an immersive analytic tool for spectral data, we examine an integrated IVR real-time model selection algorithm for a recent model updating method that adapts a model from the original calibration domain to predict samples from shifted target domains. Using near-infrared data, analyte prediction errors from IVR-selected models are reduced compared to errors using an established autonomous model selection approach. Results demonstrate the viability of IVR as a human data analysis interface for spectral data analysis including classification problems.

Diagnosing retinal disorders with artificial intelligence: the role of large language models in interpreting pattern electroretinography data

Aims: To evaluate the diagnostic accuracy of Claude-3, a large language model, in detecting pathological features and diagnosing retinitis pigmentosa and cone-rod dystrophy using pattern electroretinography data. Methods: A subset of pattern electroretinography measurements from healthy individuals, patients with retinitis pigmentosa and cone-rod dystrophy was randomly selected from the PERG-IOBA dataset. The pattern electroretinography and clinical data, including age, gender, visual acuities, were provided to Claude-3 for analysis and diagnostic predictions. The model’s accuracy was assessed in two scenarios: “first choice,” evaluating the accuracy of the primary differential diagnosis and “top 3,” evaluating whether the correct diagnosis was included within the top three differential diagnoses. Results: A total of 46 subjects were included in the study: 20 healthy individuals, 13 patients with retinitis pigmentosa, 13 patients with cone-rod dystrophy. Claude-3 achieved 100% accuracy in detecting the presence or absence of pathology. In the “first choice” scenario, the model demonstrated moderate accuracy in diagnosing retinitis pigmentosa (61.5%) and cone-rod dystrophy (53.8%). However, in the “top 3” scenario, the model’s performance significantly improved, with accuracies of 92.3% for retinitis pigmentosa and 76.9% for cone-rod dystrophy. Conclusion: This is the first study to demonstrate the potential of large language models, specifically Claude-3, in analyzing pattern electroretinography data to diagnose retinal disorders. Despite some limitations, the model’s high accuracy in detecting pathologies and distinguishing between specific diseases highlights the potential of large language models in ocular electrophysiology. Future research should focus on integrating multimodal data, and conducting comparative analyses with human experts.

Dataset from a human-in-the-loop approach to identify functionally important protein residues from literature

We present a novel system that leverages curators in the loop to develop a dataset and model for detecting structure features and functional annotations at residue-level from standard publication text. Our approach involves the integration of data from multiple resources, including PDBe, EuropePMC, PubMedCentral, and PubMed, combined with annotation guidelines from UniProt, and LitSuggest and HuggingFace models as tools in the annotation process. A team of seven annotators manually curated ten articles for named entities, which we utilized to train a starting PubmedBert model from HuggingFace. Using a human-in-the-loop annotation system, we iteratively developed the best model with commendable performance metrics of 0.90 for precision, 0.92 for recall, and 0.91 for F1-measure. Our proposed system showcases a successful synergy of machine learning techniques and human expertise in curating a dataset for residue-level functional annotations and protein structure features. The results demonstrate the potential for broader applications in protein research, bridging the gap between advanced machine learning models and the indispensable insights of domain experts.

Multi-Split Configuration Design for Fluid-Based Thermal Management Systems

Abstract High power density systems require efficient cooling to maintain their thermal performance. Despite this, as systems get larger and more complex, human expertise and insight may not suffice to determine the desired thermal management system designs. To this end, a framework for automatic architecture exploration is presented in this article for a class of single-phase, multi-split cooling systems. For this class of systems, heat generation devices are clustered based on their spatial information, and flow splits are added only when required and at the location of heat devices. To generate different architectures, candidate architectures are represented as graphs. From these graphs, dynamic physics models are created automatically using a graph-based thermal modeling framework. Then, an optimal fluid flow distribution problem is solved by addressing temperature constraints in the presence of exogenous heat loads to achieve optimal performance. The focus in this work is on the design of general multi-split heat management systems. The methods presented here can be used for diverse applications in the domain of configuration design. The multi-split algorithm can produce configurations where splitting can occur at any of the vertices. The results presented include three categories of problems and are discussed in detail.

A foundation model for clinician-centered drug repurposing.

Drug repurposing-identifying new therapeutic uses for approved drugs-is often a serendipitous and opportunistic endeavour to expand the use of drugs for new diseases. The clinical utility of drug-repurposing artificial intelligence (AI) models remains limited because these models focus narrowly on diseases for which some drugs already exist. Here we introduce TxGNN, a graph foundation model for zero-shot drug repurposing, identifying therapeutic candidates even for diseases with limited treatment options or no existing drugs. Trained on a medical knowledge graph, TxGNN uses a graph neural network and metric learning module to rank drugs as potential indications and contraindications for 17,080 diseases. When benchmarked against 8 methods, TxGNN improves prediction accuracy for indications by 49.2% and contraindications by 35.1% under stringent zero-shot evaluation. To facilitate model interpretation, TxGNN's Explainer module offers transparent insights into multi-hop medical knowledge paths that form TxGNN's predictive rationales. Human evaluation of TxGNN's Explainer showed that TxGNN's predictions and explanations perform encouragingly on multiple axes of performance beyond accuracy. Many of TxGNN's new predictions align well with off-label prescriptions that clinicians previously made in a large healthcare system. TxGNN's drug-repurposing predictions are accurate, consistent with off-label drug use, and can be investigated by human experts through multi-hop interpretable rationales.

Between Two Worlds: Investigating the Intersection of Human Expertise and Machine Learning in the Case of Coronary Artery Disease Diagnosis.

Coronary artery disease (CAD) presents a significant global health burden, with early and accurate diagnostics crucial for effective management and treatment strategies. This study evaluates the efficacy of human evaluators compared to a Random Forest (RF) machine learning model in predicting CAD risk. It investigates the impact of incorporating human clinical judgments into the RF model's predictive capabilities. We recruited 606 patients from the Department of Nuclear Medicine at the University Hospital of Patras, Greece, from 16 February 2018 to 28 February 2022. Clinical data inputs included age, sex, comprehensive cardiovascular history (including prior myocardial infarction and revascularisation), CAD predisposing factors (such as hypertension, dyslipidemia, smoking, diabetes, and peripheral arteriopathy), baseline ECG abnormalities, and symptomatic descriptions ranging from asymptomatic states to angina-like symptoms and dyspnea on exertion. The diagnostic accuracies of human evaluators and the RF model (when trained with datasets inclusive of human judges' assessments) were comparable at 79% and 80.17%, respectively. However, the performance of the RF model notably declined to 73.76% when human clinical judgments were excluded from its training dataset. These results highlight a potential synergistic relationship between human expertise and advanced algorithmic predictions, suggesting a hybrid approach as a promising direction for enhancing CAD diagnostics.

Representation Learning for Sequential Volumetric Design Tasks

Abstract Volumetric design, also called massing design, is the first and critical step in professional building design which is sequential in nature. As the volumetric design process is complex, the underlying sequential design process en- codes valuable information for designers. Many efforts have been made to automatically generate reasonable volumetric designs, but the quality of the generated design solutions varies, and evaluating a design solution requires either a prohibitively comprehensive set of metrics or expensive human expertise. While previous approaches focused on learning only the final design instead of sequential design tasks, we propose to encode the design knowledge from a collection of expert or high-performing design sequences and extract useful representations using transformer-based models. Later we propose to utilize the learned representations for crucial downstream applications such as design preference evaluation and procedural design generation. We develop the preference model by estimating the density of the learned representations whereas we train an autoregressive transformer model for sequential design generation. We demonstrate our ideas by leveraging a novel dataset of thousands of sequential volumetric designs. Our preference model can compare two arbitrarily given design sequences and is almost 90% accurate in evaluation against random design sequences. Our autoregressive model is also capable of autocompleting a volumetric design sequence from a partial design sequence.

Use of deep learning to improve seismic data quality analysis

Seismic data are influenced by various types of noise, which are typically categorized into two primary classes: anthropogenic and environmental. However, the detection of instrumental noise or malfunctioning stations also plays a crucial role in ensuring the data quality and the efficiency of a seismic network. The visual inspection of seismic spectral diagrams (e.g. power spectral density) enables us to identify issues that could potentially compromise data quality, thereby affecting subsequent calculations such as Magnitude or Peak Ground Acceleration (PGA). However, this process is time-consuming and demands significant human expertise due to the complexity of the diagrams, compounded by the sheer number of stations requiring examination. Therefore, in this paper, we explore the feasibility of transferring human expertise into an artificial intelligence system to create an automated system capable of rapidly performing such detection. More specifically, in the first part of this paper, we use Probability Density Function (PDF) diagrams, enabling an initial assessment of station performance via visual inspection. We describe this plot type and provide examples that reveal whether a station is functioning correctly or if technical issues exist. A table containing the main evaluation criteria is provided. In the second part of this paper, we demonstrate that these plots can serve as input for a neural network, allowing the development of the aforementioned automated system. Through extensive testing under various conditions, we have observed that the trained network consistently achieves an accuracy rate exceeding 85% across all four conductedtests. In the latest and most significant test, the achieved accuracy is approximately 87%.

3-1-3 Weight averaging technique-based performance evaluation of deep neural networks for Alzheimer’s disease detection using structural MRI

Alzheimer’s disease (AD) is a progressive neurological disorder. It is identified by the gradual shrinkage of the brain and the loss of brain cells. This leads to cognitive decline and impaired social functioning, making it a major contributor to dementia. While there are no treatments to reverse AD’s progression, spotting the disease’s onset can have a significant impact in the medical field. Deep learning (DL) has revolutionized medical image classification by automating feature engineering, removing the requirement for human experts in feature extraction. DL-based solutions are highly accurate but demand a lot of training data, which poses a common challenge. Transfer learning (TL) has gained attention for its knack for handling limited data and expediting model training. This study uses TL to classify AD using T1-weighted 3D Magnetic Resonance Imaging (MRI) from the Alzheimer’s Disease Neuroimaging (ADNI) database. Four modified pre-trained deep neural networks (DNN), VGG16, MobileNet, DenseNet121, and NASNetMobile, are trained and evaluated on the ADNI dataset. The 3-1-3 weight averaging technique and fine-tuning improve the performance of the classification models. The evaluated accuracies for AD classification are VGG16: 98.75%; MobileNet: 97.5%; DenseNet: 97.5%; and NASNetMobile: 96.25%. The receiver operating characteristic (ROC), precision–recall (PR), and Kolmogorov-Smirnov (KS) statistic plots validate the effectiveness of the modified pre-trained model. Modified VGG16 excels with area under the curve (AUC) values of 0.99 for ROC and 0.998 for PR curves. The proposed approach shows effective AD classification by achieving high accuracy using the 3-1-3 weight averaging technique and fine-tuning.