تطبيقات التعلم العميق في المعلوماتية الحيوية والطبية المعتمدة على إنترنت الأشياء: مراجعة منهجية للأدبيات
The deep learning applications in IoT-based bio- and medical informatics: a systematic literature review

المجلة: Neural Computing and Applications، المجلد: 36، العدد: 11
DOI: https://doi.org/10.1007/s00521-023-09366-3
تاريخ النشر: 2024-01-13
المؤلف: Zahra Mohtasham‐Amiri وآخرون
الموضوع الرئيسي: التقنيات المتقدمة في مجالات متنوعة

نظرة عامة

تقدم الورقة البحثية نظرة شاملة على تطبيق تقنيات التعلم العميق (DL) في مجال المعلوماتية الحيوية والمعلوماتية الطبية المعتمدة على إنترنت الأشياء (IoT). تبرز الأهمية المتزايدة للتعلم الآلي (ML) والتعلم العميق (DL) في تحسين نتائج الرعاية الصحية من خلال التحليل الفوري للبيانات المعقدة. يصنف المؤلفون أحدث حلول DL إلى خمسة أنواع متميزة: الشبكات العصبية التلافيفية، الشبكات العصبية المتكررة، الشبكات التنافسية التوليدية، البيرسيبترونات متعددة الطبقات، والأساليب الهجينة. يتم إجراء مراجعة منهجية للأدبيات لتقييم هذه التقنيات بناءً على مقاييس الأداء المختلفة، بما في ذلك الدقة، الحساسية، الخصوصية، F-score، الكمون، القابلية للتكيف، وقابلية التوسع، مع معالجة التحديات التي تواجه تنفيذ DL في هذا المجال.

تؤكد الخاتمة على التقدم الملحوظ في تطبيقات DL لاكتشاف الأدوية، وتشخيص الأمراض، ومراقبة المرضى، مع التأكيد على الحاجة إلى البحث المستمر لتحسين هذه المنهجيات وزيادة قوتها في بيئات الرعاية الصحية الواقعية. يشير المؤلفون إلى أنه بينما تركز العديد من الدراسات على مؤشرات الأداء الرئيسية، يتم تجاهل جوانب مثل الأمان ووقت التقارب في كثير من الأحيان. تشمل قيود المراجعة استبعاد الأوراق غير الإنجليزية وعدم وضوح بعض أوصاف الخوارزميات، مما قد يعيق الفهم الشامل للمجال. تهدف الورقة إلى أن تكون مصدرًا قيمًا للبحث المستقبلي في تطبيقات DL ضمن المعلوماتية الطبية والحيوية.

مقدمة

تستعرض مقدمة الورقة الدور الحاسم للمعلوماتية الحيوية كحقل متعدد التخصصات يدمج علوم الحاسوب، وعلم الأحياء، وتحليل البيانات لإدارة وتفسير مجموعات البيانات البيولوجية الضخمة، لا سيما في علم الجينوم. تؤكد على ضرورة المعلوماتية الحيوية في الطب الحديث من أجل التعامل الفعال مع البيانات وتطبيق الحوسبة الموزعة لمعالجة المعلومات الجينومية على نطاق واسع. تبرز الورقة الأهمية المتزايدة لتقنيات التعلم الآلي (ML) والتعلم العميق (DL) في المعلوماتية الحيوية، والتي تسهل تحليل البيانات البيولوجية المعقدة، مما يمكّن من الحصول على رؤى حول الطفرات الجينية وتطوير علاجات جديدة.

تهدف الدراسة إلى تقديم نظرة شاملة على تطبيقات ML في المعلوماتية الطبية والحيوية المعتمدة على إنترنت الأشياء (IoT)، وتصنيف استراتيجيات DL إلى خمس مجموعات: الشبكات العصبية التلافيفية (CNN)، الشبكات العصبية المتكررة (RNN)، الشبكات التنافسية التوليدية (GAN)، البيرسيبترونات متعددة الطبقات (MLP)، والأساليب الهجينة. تجري الدراسة مراجعة منهجية للأدبيات (SLR) لتحليل ومقارنة هذه الآليات DL، مناقشة خصائصها، ومنهجياتها، وتطبيقاتها مع تحديد اتجاهات البحث المستقبلية والتحديات الحالية. تشمل مساهمات الورقة تقييمًا شاملاً لأساليب ML/DL الحالية، ورؤى حول تطبيقاتها في المعلوماتية الحيوية، وإطار عمل لمعالجة التطورات المستقبلية في هذا المجال.

الطرق

تتضمن المنهجية المستخدمة في هذا البحث مراجعة منهجية للأدبيات (SLR) لتقييم تطبيق التعلم الآلي (ML) في المعلوماتية الحيوية والطبية بشكل شامل. تسهل هذه الطريقة فحصًا دقيقًا لآليات التعلم العميق (DL) وتقييم صلاحية استراتيجية اختيار الدراسة. تحدد الأقسام الفرعية التالية عملية التحقيق، بما في ذلك صياغة أسئلة البحث ومعايير اختيار الأوراق.

يمثل دمج تطبيقات إنترنت الأشياء (IoT) مع طرق DL تقدمًا كبيرًا في تكنولوجيا الرعاية الصحية. تستخدم هذه التطبيقات الطبيعة المترابطة للأجهزة وأجهزة الاستشعار ضمن إطار عمل IoT لتعزيز رعاية المرضى، والتشخيص، والعلاج. تعتبر خوارزميات DL بارعة في معالجة مجموعات البيانات الكبيرة والمعقدة، مما يسمح بتحليل تدفقات البيانات البيولوجية المتنوعة مثل القياسات الفسيولوجية، والتصوير الطبي، والبيانات الجينومية. تدعم هذه القدرة المراقبة الصحية في الوقت الحقيقي، والكشف المبكر عن الشذوذ، وتطوير خطط علاج شخصية. علاوة على ذلك، تساهم النماذج التنبؤية المعتمدة على DL في تقييمات تنبؤية دقيقة وتقدم الطب الدقيق. لا يحسن دمج تقنيات DL وIoT أمان البيانات وخصوصيتها فحسب، مما يضمن الامتثال للوائح الرعاية الصحية، بل لديه أيضًا القدرة على رفع جودة تقديم الرعاية الصحية بشكل كبير وتعزيز الابتكارات التي يمكن أن تحول الممارسة الطبية.

النتائج

يوفر قسم النتائج في الدراسة تحليلًا شاملاً لتقنيات التعلم العميق (DL) والتعلم الآلي (ML) في المعلوماتية الحيوية والمعلوماتية الطبية، مع تسليط الضوء على تطبيقاتها وتحدياتها. تكشف التحقيقات أنه بينما يتم اعتماد طرق DL، لا سيما الشبكات العصبية التلافيفية (CNNs)، والشبكات العصبية المتكررة (RNNs)، والشبكات التنافسية التوليدية (GANs)، على نطاق واسع لمهام مثل تصنيف الصور والكشف عن الشذوذ، هناك قيود كبيرة بسبب ندرة مجموعات البيانات الكبيرة وعالية الجودة. يؤكد المؤلفون على أهمية هيكلة مدخلات البيانات ومهام التدريب بشكل فعال لتعزيز أداء ML وتشجيع المزيد من الاستكشاف في هذه المجالات.

كما تحدد التحليلات الاتجاهات في مساهمات البحث، حيث تظهر الصين كأحد المساهمين الرئيسيين في هذا المجال. تعتبر Python، إلى جانب مكتبتها TensorFlow، هي البيئة البرمجية السائدة المستخدمة لتنفيذ نماذج DL، مما يعكس مرونتها ومواردها الواسعة. تشير الدراسة إلى أنه بينما تعتبر الدقة محور التركيز الرئيسي في تقييم أداء النموذج، غالبًا ما يتم تجاهل الدقة، مما قد يؤدي إلى تصنيفات خاطئة حرجة في التطبيقات الطبية. يدعو المؤلفون إلى نهج أكثر شمولية في تقييم النماذج من خلال دمج مقاييس متعددة، مثل الحساسية والخصوصية، لضمان فهم شامل لفعالية النموذج في السيناريوهات الواقعية. بشكل عام، تؤكد النتائج على الانتشار المتزايد لتقنيات DL في الرعاية الصحية والمعلوماتية الحيوية، بينما تبرز أيضًا الحاجة إلى تحسين تكامل البيانات وممارسات التقييم.

المناقشة

يوفر قسم المناقشة في الورقة نظرة شاملة على منهجيات التعلم العميق (DL) وتطبيقاتها في المعلوماتية الحيوية والمعلوماتية الطبية. يصنف أساليب DL إلى ثلاثة أنواع رئيسية: التعلم تحت الإشراف، التعلم شبه المراقب، والتعلم غير المراقب. يتم تسليط الضوء على التعلم تحت الإشراف لفعاليته في مهام مثل التعرف على الصور وكشف البريد العشوائي، بينما يعزز التعلم شبه المراقب الدقة من خلال الاستفادة من البيانات المعلّمة وغير المعلّمة. يركز التعلم غير المراقب على تحديد الأنماط والعلاقات داخل البيانات، وهو مفيد بشكل خاص في الكشف عن الشذوذ واستخراج الميزات. يؤكد القسم على الإمكانات التحويلية لـ DL في المعلوماتية الحيوية، حيث يساعد في علم الجينوم، ونمذجة البروتينات، وتصميم الأدوية، مما يعزز في النهاية فهمنا للعمليات البيولوجية.

تتناول الورقة أيضًا تطبيقات محددة لتقنيات DL في المعلوماتية الحيوية، بما في ذلك الكشف عن الإنزيمات باستخدام الشبكات العصبية متعددة الطبقات، والانحدار في التعبير الجيني، وتوقع تفاعلات RNA-بروتين باستخدام الشبكات العصبية التلافيفية (CNNs). تناقش استخدام الشبكات العصبية المتكررة (RNNs) وCNNs لتوقع أداء تسلسل الحمض النووي، بالإضافة إلى تطبيق الشبكات المتبقية (ResNet) والتعلم الانتقالي في تصنيف الصور الطبية. بالإضافة إلى ذلك، تستكشف استخدام الشبكات التلافيفية البيانية (GCNs) لتوقع تفاعلات البروتينات والشبكات التنافسية التوليدية (GANs) لتحسين دقة الصور البيولوجية. يختتم القسم بالتأكيد على الحاجة إلى مزيد من الاستكشاف لقدرات DL في معالجة التحديات في المعلوماتية الحيوية والمعلوماتية الطبية، داعيًا إلى استمرار البحث لتحسين نتائج المرضى وتقدم الطب الدقيق.

Journal: Neural Computing and Applications, Volume: 36, Issue: 11
DOI: https://doi.org/10.1007/s00521-023-09366-3
Publication Date: 2024-01-13
Author(s): Zahra Mohtasham‐Amiri et al.
Primary Topic: Advanced Technologies in Various Fields

Overview

The research paper provides a comprehensive overview of the application of deep learning (DL) techniques in the realm of Internet of Things (IoT)-based bioinformatics and medical informatics. It highlights the growing significance of ML and DL in enhancing healthcare outcomes through real-time analysis of complex data. The authors categorize the latest DL solutions into five distinct types: convolutional neural networks, recurrent neural networks, generative adversarial networks, multilayer perceptrons, and hybrid methods. A systematic literature review is conducted to evaluate these techniques based on various performance metrics, including accuracy, sensitivity, specificity, F-score, latency, adaptability, and scalability, while also addressing the challenges faced in implementing DL in this field.

The conclusion emphasizes the remarkable advancements in DL applications for drug discovery, disease diagnosis, and patient monitoring, while underscoring the need for ongoing research to refine these methodologies and improve their robustness in real-world healthcare settings. The authors note that while many studies focus on key performance indicators, aspects such as security and convergence time are often overlooked. Limitations of the review include the exclusion of non-English papers and insufficient clarity in some algorithm descriptions, which may hinder a comprehensive understanding of the field. The paper aims to serve as a valuable resource for future research in DL applications within medical and bioinformatics.

Introduction

The introduction of the paper outlines the critical role of bioinformatics as an interdisciplinary field that integrates computer science, biology, and data analysis to manage and interpret vast biological datasets, particularly in genomics. It emphasizes the necessity of bioinformatics in modern medicine for efficient data handling and the application of distributed computing to process large-scale genomic information. The paper highlights the growing importance of machine learning (ML) and deep learning (DL) techniques in bioinformatics, which facilitate the analysis of complex biological data, enabling insights into genetic mutations and the development of new treatments.

The research aims to provide a comprehensive overview of ML applications in Internet of Things (IoT)-based medical and bioinformatics, categorizing DL strategies into five groups: convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), multilayer perceptrons (MLP), and hybrid approaches. The study conducts a systematic literature review (SLR) to analyze and compare these DL mechanisms, discussing their properties, methodologies, and applications while identifying future research directions and existing challenges. The paper’s contributions include a thorough evaluation of current ML/DL methods, insights into their applications in bioinformatics, and a framework for addressing future developments in the field.

Methods

The methodology employed in this research involves a systematic literature review (SLR) to comprehensively assess the application of machine learning (ML) in bio-and medical informatics. This approach facilitates a detailed examination of deep learning (DL) mechanisms and evaluates the validity of the study selection strategy. Subsequent subsections outline the investigation process, including the formulation of research questions and criteria for paper selection.

The integration of Internet of Things (IoT) applications with DL methods represents a significant advancement in healthcare technology. These applications utilize the interconnected nature of devices and sensors within the IoT framework to enhance patient care, diagnosis, and treatment. DL algorithms are adept at processing large and complex datasets, allowing for the analysis of various biomedical data streams such as physiological measurements, medical imaging, and genomic data. This capability supports real-time health monitoring, early anomaly detection, and the development of personalized treatment plans. Furthermore, DL-based predictive models contribute to accurate prognostic assessments and the advancement of precision medicine. The combination of DL and IoT technologies not only improves data security and privacy, ensuring adherence to healthcare regulations, but also has the potential to significantly elevate the quality of healthcare delivery and foster innovations that could transform medical practice.

Results

The results section of the study provides a comprehensive analysis of deep learning (DL) and machine learning (ML) techniques in bioinformatics and medical informatics, highlighting their applications and challenges. The investigation reveals that while DL methods, particularly Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs), are widely adopted for tasks such as image classification and anomaly detection, there are significant limitations due to the scarcity of large, high-quality datasets. The authors emphasize the importance of structuring data inputs and training tasks effectively to enhance ML performance and encourage further exploration in these domains.

The analysis also identifies trends in research contributions, with China emerging as a leading contributor in the field. Python, along with its TensorFlow library, is the predominant programming environment used for implementing DL models, reflecting its versatility and extensive resources. The study notes that while accuracy is a primary focus in evaluating model performance, precision is often overlooked, which could lead to critical misclassifications in medical applications. The authors advocate for a more holistic approach to model evaluation by incorporating multiple metrics, such as sensitivity and specificity, to ensure a comprehensive understanding of model effectiveness in real-world scenarios. Overall, the findings underscore the growing prevalence of DL techniques in healthcare and bioinformatics, while also highlighting the need for improved data integration and evaluation practices.

Discussion

The discussion section of the paper provides a comprehensive overview of deep learning (DL) methodologies and their applications in bioinformatics and medical informatics. It categorizes DL approaches into three main types: supervised, semi-supervised, and unsupervised learning. Supervised learning is highlighted for its effectiveness in tasks such as image recognition and spam detection, while semi-supervised learning enhances accuracy by leveraging both labeled and unlabeled data. Unsupervised learning focuses on identifying patterns and relationships within data, which is particularly useful in anomaly detection and feature extraction. The section emphasizes the transformative potential of DL in bioinformatics, where it aids in genomics, protein modeling, and drug design, ultimately enhancing our understanding of biological processes.

The paper further details specific applications of DL techniques in bioinformatics, including enzyme detection using multilayer neural networks, gene expression regression, and predicting RNA-protein interactions with convolutional neural networks (CNNs). It discusses the use of recurrent neural networks (RNNs) and CNNs for DNA sequence performance prediction, as well as the application of Residual Networks (ResNet) and transfer learning in biomedical image classification. Additionally, it explores the use of Graph Convolutional Networks (GCNs) for protein interaction prediction and Generative Adversarial Networks (GANs) for enhancing biological image resolution. The section concludes by underscoring the need for further exploration of DL’s capabilities in addressing challenges in bioinformatics and medical informatics, advocating for continued research to improve patient outcomes and advance precision medicine.