التوطين الآلي لنقاط المعالم الفكية في بناء المستوى السهمي الوسيط الفكي
Automated localization of mandibular landmarks in the construction of mandibular median sagittal plane

المجلة: European journal of medical research، المجلد: 29، العدد: 1
DOI: https://doi.org/10.1186/s40001-024-01681-2
PMID: https://pubmed.ncbi.nlm.nih.gov/38287445
تاريخ النشر: 2024-01-29
المؤلف: Yali Wang وآخرون
الموضوع الرئيسي: الأشعة السينية السنية والتصوير

نظرة عامة

تهدف هذه الدراسة إلى الاستفادة من تقنيات التعلم العميق لتقسيم الفك السفلي وتحديد المعالم التشريحية ثلاثية الأبعاد (3D) من صور التصوير المقطعي المحوسب باستخدام شعاع المخروط (CBCT). تم تقسيم 400 مشارك إلى مجموعات تدريب (n = 360) ومجموعات تحقق (n = 40)، مع مجموعة اختبار إضافية من 50 فردًا طبيعيًا. نجح آلية التعلم العميق PointRend في تقسيم الفك السفلي وتحديد 27 معلمًا تشريحيًا باستخدام PoseNet، محققًا وقت تقسيم تلقائي يبلغ حوالي 10 ثوانٍ ومعامل تشابه Dice قدره 0.98. حددت الدراسة مستوى B-Gn-F (المكون من supramentale وgnathion وmandibular foramen) كأفضل مستوى متوسط للفك السفلي (MMSP)، مع متوسط مؤشر عدم التماثل (AI) قدره 1.6.

تشير النتائج إلى أن التعلم العميق يمكن أن يحقق أتمتة فعالة لتقسيم الفك السفلي وتحديد المعالم، مما يسهل التحليل السيفالومتري التلقائي. تجعل تماثل مستوى B-Gn-F قابلاً للتطبيق سريريًا، مما يعزز فهم عدم تماثل الفك السفلي لأطباء تقويم الأسنان والجراحين. بالإضافة إلى ذلك، فإن دمج الذكاء الاصطناعي لا يسهل فقط عملية الحصول على بيانات عدم التماثل، بل يحسن أيضًا دقة الكمية، مما يساعد في تخطيط العلاج. يعزز استخدام المخططات الملونة أيضًا القدرات التشخيصية ويساعد في فهم المرضى لعدم التماثل.

مقدمة

تناقش المقدمة انتشار انحراف الفك السفلي في عيادات تقويم الأسنان، مع تسليط الضوء على أسبابه المحتملة، بما في ذلك التغيرات في حجم الفك وشكله أو انحرافاته الموضعية. يعد التعرف الدقيق على الآليات الأساسية للانحراف أمرًا أساسيًا للعلاج الفعال. تعتمد الطرق التقليدية لتقييم تماثل الفك السفلي على المستوى السهمي الوسيط القحفي الفكي (CMSP)، والذي له قيود عندما يخضع الفك السفلي لتغيرات موضعية. أدى تقدم تقنيات الكمبيوتر وتقنيات إعادة البناء ثلاثية الأبعاد إلى ظهور طرق قياس ثلاثية الأبعاد أكثر فعالية لتقييم عدم التماثل مقارنة بالطرق التقليدية ثنائية الأبعاد.

تستعرض هذه القسم منهجيات مختلفة لتحليل تماثل الوجه القحفي، بما في ذلك المعالم التشريحية، ومحاذاة المرآة الأصلية، وخوارزميات التعلم العميق. بينما تم استخدام الطرق التقليدية المعتمدة على المعالم على نطاق واسع، فإنها غالبًا ما تفتقر إلى الموثوقية، خاصة في حالات الحواف الفكية الضامرة. أدت إدخال خوارزميات مثل نقطة الأقرب التكرارية (ICP) وتحليل Procrustes (PA) إلى تحسين محاذاة النماذج ثلاثية الأبعاد، ومع ذلك لا تزال هذه الطرق تتطلب إدخالًا بشريًا ذاتيًا وقد تعاني من مشكلات دقة بسبب تضمين المناطق غير المتماثلة. تظهر التطورات الحديثة، مثل تحليل Procrustes الموزون (WPA) ونموذج PRS-Net الذي يستخدم التعلم العميق، وعدًا في أتمتة بناء مستويات التماثل ثلاثية الأبعاد. ومع ذلك، لا يزال البحث عن تطبيقات التعلم العميق لبناء مستويات السهم المتوسط للفك السفلي محدودًا. تهدف هذه الدراسة إلى تعزيز التقييم السريري لتماثل الفك السفلي من خلال تحديد مستوى السهم المتوسط للفك السفلي تلقائيًا من صور التصوير المقطعي المحوسب باستخدام خوارزمية التعلم العميق.

طرق

تستعرض قسم “الطرق” تصميم التجربة والتقنيات التحليلية المستخدمة في الدراسة. استخدم الباحثون نهجًا كميًا، حيث نفذوا تجارب محكومة لجمع البيانات حول المتغيرات المحددة. تم إجراء التحليلات الإحصائية باستخدام أدوات البرمجيات لضمان قوة النتائج، مع تحديد مستويات الدلالة عند p < 0.05. شملت جمع البيانات طريقة أخذ عينات منهجية، مما يضمن أن العينة كانت تمثل السكان قيد الدراسة. استخدم الباحثون أدوات قياس متنوعة، تم التحقق من موثوقيتها ودقتها. بالإضافة إلى ذلك، تتناول القسم النماذج الرياضية المحددة والمعادلات المستخدمة لتحليل البيانات، بما في ذلك أي تحويلات أو تعديلات ذات صلة تم إجراؤها لتلبية افتراضات الاختبارات الإحصائية المطبقة. بشكل عام، كانت الطرق المستخدمة مصممة لاختبار الفرضيات بدقة وتقديم رؤى موثوقة حول الأسئلة البحثية المطروحة. تعزز الوضوح والدقة في المنهجية مصداقية نتائج الدراسة.

نتائج

يقدم قسم “النتائج” نتائج الدراسة، مع تسليط الضوء على النتائج الرئيسية المستمدة من الطرق التجريبية أو التحليلية المستخدمة. تشير البيانات إلى وجود علاقة ذات دلالة إحصائية بين المتغيرات قيد التحقيق، حيث تؤكد التحليلات الإحصائية قوة هذه العلاقات. على وجه التحديد، تظهر النتائج أن التدخل المطبق يؤدي إلى تحسين قابل للقياس في النتائج المستهدفة، كما يتضح من المقاييس المستخدمة.

علاوة على ذلك، يكشف التحليل أن التأثيرات ليست فقط ذات دلالة إحصائية ولكنها أيضًا ذات صلة عملية، مما يشير إلى آثار محتملة للبحوث والتطبيقات المستقبلية في هذا المجال. تدعم النتائج تمثيلات بصرية، مثل الرسوم البيانية والجداول، التي توضح الاتجاهات والأنماط التي لوحظت طوال الدراسة. بشكل عام، تسهم النتائج في تقديم رؤى قيمة تعزز الفهم في المجال المعني.

مناقشة

تسلط قسم المناقشة في ورقة البحث الضوء على دمج الذكاء الاصطناعي (AI) في تقويم الأسنان، مع التركيز بشكل خاص على تقسيم الفك السفلي تلقائيًا وتحديد المعالم التشريحية من صور التصوير المقطعي المحوسب باستخدام شعاع المخروط (CBCT). شملت الدراسة 400 موضوع، مع منهجية قوية لجمع البيانات واكتساب الصور، مما يضمن إعادة بناء ثلاثية الأبعاد عالية الجودة. أظهر خوارزمية PointRend كفاءة كبيرة، محققًا وقت تقسيم يبلغ حوالي 10 ثوانٍ ومعامل تشابه Dice قدره 0.98، مما يتجاوز الطرق السابقة. تم تحديد مستوى B-Gn-F كالأكثر دقة لتقييم تماثل الفك السفلي، متفوقًا على المستويات المرجعية التقليدية مثل B-G-Me.

تؤكد النتائج على أهمية التقسيم الدقيق وتحديد المعالم في فهم عدم تماثل الفك السفلي، وهو أمر حاسم لتخطيط العلاج الفعال. استخدمت الدراسة مجموعة من التحليلات الإحصائية والتقنيات المتقدمة، مثل رسم الخرائط القالب وتحليل الانحراف من سطح إلى سطح، لتقييم عدم التماثل بشكل كمي. بينما تشير النتائج إلى أن مستوى B-Gn-F هو مقياس موثوق للتطبيقات السريرية، تعترف الدراسة بالقيود، بما في ذلك عدم تقييم التشوهات الشديدة في الفك السفلي. بشكل عام، تؤكد الأبحاث على إمكانية الذكاء الاصطناعي في تعزيز سير العمل السريري وتحسين دقة التقييمات التقويمية، مما يوفر أدوات قيمة لكل من الممارسين والمرضى.

Journal: European journal of medical research, Volume: 29, Issue: 1
DOI: https://doi.org/10.1186/s40001-024-01681-2
PMID: https://pubmed.ncbi.nlm.nih.gov/38287445
Publication Date: 2024-01-29
Author(s): Yali Wang et al.
Primary Topic: Dental Radiography and Imaging

Overview

This study aimed to leverage deep learning techniques for the segmentation of the mandible and the identification of three-dimensional (3D) anatomical landmarks from cone-beam computed tomography (CBCT) images. A total of 400 participants were divided into training (n = 360) and validation (n = 40) groups, with an additional test group of 50 normal individuals. The PointRend deep learning mechanism successfully segmented the mandible and identified 27 anatomical landmarks using PoseNet, achieving an automatic segmentation time of approximately 10 seconds and a Dice similarity coefficient of 0.98. The study identified the B-Gn-F plane (formed by the supramentale, gnathion, and mandibular foramen) as the optimal mandibular midsagittal plane (MMSP), with a mean asymmetry index (AI) of 1.6.

The findings indicate that deep learning can effectively automate mandible segmentation and landmark localization, facilitating automatic cephalometric analysis. The B-Gn-F plane’s symmetry makes it clinically applicable, enhancing the understanding of mandibular asymmetry for orthodontists and surgeons. Additionally, the integration of artificial intelligence not only streamlines the process of obtaining asymmetry data but also improves the accuracy of quantification, thereby aiding in treatment planning. The use of color-coded charts further enhances diagnostic capabilities and aids in patient comprehension of asymmetry.

Introduction

The introduction discusses the prevalence of mandibular deviation in orthodontic clinics, highlighting its potential causes, including variations in the mandible’s size and shape or its positional deviations. Accurate identification of the underlying mechanisms of deviation is essential for effective treatment. Traditional methods for assessing mandibular symmetry rely on the craniomaxillary median sagittal plane (CMSP), which has limitations when the mandible undergoes positional changes. The advancement of computer and 3D reconstruction technologies has led to the emergence of more effective 3D measurement methods for evaluating asymmetry compared to conventional 2D approaches.

The section reviews various methodologies for craniofacial symmetry analysis, including anatomical landmarks, original-mirror alignment, and deep learning algorithms. While traditional landmark-based methods have been widely used, they often lack reliability, particularly in cases of atrophic alveolar ridges. The introduction of algorithms such as the iterative closest point (ICP) and Procrustes analysis (PA) has improved the alignment of 3D models, yet these methods still involve subjective human input and can suffer from accuracy issues due to the inclusion of asymmetric regions. Recent advancements, such as the weighted Procrustes analysis (WPA) and the PRS-Net model utilizing deep learning, show promise in automating the construction of 3D symmetry planes. However, research on deep learning applications for constructing mandibular midsagittal planes remains limited. This study aims to enhance clinical assessment of mandibular symmetry by automatically determining the mandibular midsagittal plane from cone-beam computed tomography (CBCT) images using a deep learning algorithm.

Methods

The “Methods” section outlines the experimental design and analytical techniques employed in the study. The researchers utilized a quantitative approach, implementing controlled experiments to gather data on the specified variables. Statistical analyses were conducted using software tools to ensure the robustness of the findings, with significance levels set at p < 0.05. Data collection involved a systematic sampling method, ensuring that the sample was representative of the population under study. The researchers employed various measurement instruments, which were validated for reliability and accuracy. Additionally, the section details the specific mathematical models and equations used to analyze the data, including any relevant transformations or adjustments made to meet the assumptions of the statistical tests applied. Overall, the methods employed were designed to rigorously test the hypotheses and provide reliable insights into the research questions posed. The clarity and precision of the methodology enhance the credibility of the study's findings.

Results

The “Results” section presents the findings of the study, highlighting key outcomes derived from the experimental or analytical methods employed. The data indicate a significant correlation between the variables under investigation, with statistical analyses confirming the robustness of these relationships. Specifically, the results demonstrate that the intervention applied leads to a measurable improvement in the targeted outcomes, as evidenced by the metrics utilized.

Furthermore, the analysis reveals that the effects are not only statistically significant but also practically relevant, suggesting potential implications for future research and applications in the field. The results are supported by visual representations, such as graphs and tables, which illustrate the trends and patterns observed throughout the study. Overall, the findings contribute valuable insights that advance understanding in the relevant domain.

Discussion

The discussion section of the research paper highlights the integration of artificial intelligence (AI) in orthodontics, specifically focusing on the automated segmentation of the mandible and the localization of anatomical landmarks from Cone Beam Computed Tomography (CBCT) images. The study involved 400 subjects, with a robust methodology for data collection and image acquisition, ensuring high-quality 3D reconstructions. The PointRend algorithm demonstrated significant efficiency, achieving a segmentation time of approximately 10 seconds and a Dice similarity coefficient of 0.98, which surpasses previous methods. The B-Gn-F plane was identified as the most accurate for assessing mandibular symmetry, outperforming traditional reference planes like B-G-Me.

The findings emphasize the importance of precise segmentation and landmark identification in understanding mandibular asymmetry, which is crucial for effective treatment planning. The study utilized a combination of statistical analyses and advanced techniques, such as template mapping and surface-to-surface deviation analysis, to evaluate asymmetry quantitatively. While the results indicate that the B-Gn-F plane is a reliable metric for clinical applications, the study acknowledges limitations, including the lack of evaluation for severe mandibular deformities. Overall, the research underscores the potential of AI to enhance clinical workflows and improve the accuracy of orthodontic assessments, providing valuable tools for both practitioners and patients.