ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
Study Protocol
Revised

Comparative evaluation and correlation of variations in articular disc morphology as assessed by automated segmentation using deep learning on magnetic resonance imaging (MRI) images in Class II (vertical) TMD cases, Class II (horizontal) TMD cases and Class I non-TMD cases

[version 2; peer review: 1 approved]
PUBLISHED 15 Jul 2024
Author details Author details
OPEN PEER REVIEW
REVIEWER STATUS

This article is included in the Artificial Intelligence and Machine Learning gateway.

This article is included in the Datta Meghe Institute of Higher Education and Research collection.

Abstract

Introduction: Temporomandibular disorder (TMD) encompasses several clinical manifestations, which are characterized by temporomandibular joint and masticatory muscle discomfort and dysfunction (TMJ). The best imaging technique for evaluating TMJ is magnetic resonance imaging (MRI), which makes it possible to see the anatomical and pathological characteristics of every joint component. In recent years, convolutional neural networks -based deep learning algorithms have been favoured because of their outstanding capability in recognizing objects in medical images. The objective of this study is to assess, compare and co-rrelate articular disc morphology by automated segmentation using deep learning on MRI images in skeletal Class II (vertical growth pattern) TMD cases as compared to skeletal Class II (horizontal growth pattern) TMD cases and Class I non-TMD cases

Methods: Grading of skeletal Class II (vertical growth pattern) cases and skeletal Class II (horizontal growth pattern) cases based on severity of TMD will be carried out using diagnostic criteria for temporomandibular disorders. Bilateral sagittal as well as coronal MRI images will be obtained. A convolutional neural network (CNN) encoder-decoder named U-Net will be used to segment the articular disc on MRI. Understanding the nature of variations between Class I and both types of Class IIs will help orthodontists to better predict the potential risk for the development of TMDs and accordingly take precautions while doing treatment in such cases. Moreover, it can be used to automate TMD diagnosis and other smart applications.

Conclusions: This study will aid in identifying articular disc morphology on MRI. The deep learning algorithms with effective data augmentation may perform better in MRI readings than human clinicians when using the same data, which will be advantageous for TMD diagnosis.

Keywords

temporomandibular disorder, TMD, articular disc, Class II TMD cases, automated segmentation, deep neural network, convolutional neural network

Revised Amendments from Version 1

We have done the necessary changes in this new version of the manuscript like improved the introduction part by classifying TMD and the overlapping symptoms and also the recent literature on conservative approaches for the treatment of TMD.

See the authors' detailed response to the review by Martina Ferrillo

Introduction

Background and rationale

Temporomandibular disorder (TMD) encompasses several clinical manifestations which is characterized by temporomandibular joint (TMJ) and masticatory muscle discomfort and dysfunction.1 TMD is frequently characterized by facial and pre-auricular regional pain, malocclusion, limited jaw movement, and clicking and locking of the TMJ.2

TMD patients may present overlapping symptoms with other chronic pain conditions, including headache, fibromylagia, and neurological conditions, probably through the phenomenon of central sensitization (mainly allodynia and hyperalgesia).

The expected incidence rate of TMD at first onset is 3.9%, with mild to severe pain and impairment, based on an American prospective cohort study of the general public. In industrialized nations, it affects 5–12% of the population.3 As reported previously in the literature, TMD is more common in cases with skeletal Class II (vertical growth pattern) followed by skeletal Class II (horizontal growth pattern) as compared to Class I cases. This was first reported by Pancherz in 1999.4

TMD could be divided in Group I: muscle disorders (including myofascial pain with and without mouth opening limitation); Group II: including disc displacement with or without reduction and mouth opening limitation; Group III: arthralgia, arthritis, and arthrosis.5

Non-invasive treatments, including behavioral therapy, physical therapy, drugs, occlusal splints, and laser therapy, are commonly recommended as the initial approach to improve joint range of motion, reduce pain, and prevent further degeneration. These methods are considered first-line treatments for various joint conditions. In contrast, minimally invasive techniques, such as intra-articular (IA) injections of sodium hyaluronate (including hyaluronic acid or HA), have gained attention for their potential effectiveness. IA injections, particularly when combined with joint arthrocentesis and lavage, are being explored as a viable treatment option.6

Laser acupuncture therapy (LAT) has been proposed as a needle-free alternative to traditional acupuncture, utilizing low-intensity laser light to stimulate traditional acupuncture points. This approach is characterized by its simplicity, lack of invasiveness, painlessness, and inherent safety compared to needle acupuncture therapy.7

The best imaging technique for evaluating TMJ is magnetic resonance imaging (MRI), which makes it possible to see both the anatomical and pathological characteristics of every joint component. The shape and location of the articular disc, the mandibular condyle’s shape and surface characteristics, atypical bone marrow signal in the mandible and temporal bone can all be assessed using an MRI.5 Disc displacement or deformation, an intracapsular condition affecting the disc-condylar complex, is one of the most prominent subgroups in individuals with TMD who have articular abnormality, with an incidence of 30–60%. An MRI is necessary for assessing variations in articular disc morphology and to predict the treatment outcome.3

Artificial intelligence (AI)-based dental applications have been researched to simplify dental care and enhance the health of more cases at a cheap cost and are attracting interest in a variety of clinical fields. By implementing AI-based dental applications, dental professionals can reduce the amount of time they spend performing regular tasks, which will enable them to provide more individualized, preventive, and collaborative dental care. In recent years, convolutional neural networks (CNN)-based deep learning algorithms have been more favoured because of their outstanding capabilities to recognize objects in medical images. Furthermore, as computational power has increased and open-source frameworks have become more common, CNN development has been dramatically facilitated. As a result, for detection and segmentation purposes, deep learning has been extensively used, showing encouraging results. There have been reports of a number of deep neural network topologies, including the CNN-derived fully convolutional network’s U-Net and SegNet variants.8

Clinicians and radiologists will be able to save time and interpret pictures more accurately if they can correctly identify the TMJ area’s major structures on MRI imaging. It can also be the basis for a lot of clever applications, including automatically diagnosing TMDs. It is without doubt important to accomplish this goal to automatically segregate TMJ structures from MRI images. Using these deep learning techniques, the mandibular condyle, articular eminence, and TMJ articular disc all are automatically detected. The TMJ segmentation of anatomical features in MRI volumes is made innovative in the current diagnostic image analysis investigation employing CNN-based DL methods.

Objectives

  • 1. To evaluate articular disc morphology in Class I (Non-TMD) cases, skeletal Class II (vertical growth pattern) TMD cases, skeletal Class II (horizontal growth pattern) TMD cases.

  • 2. To compare articular disc morphology in skeletal Class II (vertical growth pattern) TMD cases with skeletal Class II (horizontal growth pattern) TMD cases and Class I (Non-TMD) cases.

  • 3. To correlate the variation in articular disc morphology with skeletal pattern.

Protocol

Study design and setting

An observational and analytical study will be conducted at Sharad Pawar Dental College, Datta Meghe Institute of Higher Education and Research (DMIHER), Sawangi, Wardha, Maharashtra in collaboration with the Faculty of Engineering and Technology (FEAT), DMIHER and the Department of Radiology, Acharya Vinobha Bhave Rural Hospital (AVBRH), Sawangi, Wardha.

A total of 90 adult cases (Class I, Class II (vertical and horizontal growth pattern)) will be chosen from the Sharad Pawar Dental College’s outpatient department (OPD) of orthodontics and dentofacial orthopaedics in Sawangi, Wardha.

Participants

Inclusion criteria

  • Class I malocclusion cases.

  • Skeletal Class II (vertical growth pattern) and TMD cases.

  • Cases with skeletal Class II (horizontal growth pattern) and TMD.

  • Cases with permanent dentition.

  • Older than 18 years of age.

  • The cases will be classified into Class I and Class II (vertical and horizontal) by cephalometric measurements as shown in the Table 1.

Table 1. Cephalometric parameters for case selection of Class I and Class II (vertical and horizontal growth).

Cephalometric measurementsClass IClass II (vertical)Class II (horizontal)
ANB angle>2°>2°
Wits appraisal0–1 mm<2 mm<2 mm
Frankfort mandibular plane angle22–28°>30°<20°
Beta angle27–33°<25°<25°
Mandibular plane angle17–28°>28°<28°

Exclusion criteria

  • Skeletal Class II (vertical growth pattern) non-TMD cases.

  • Non-TMD Class II (horizontal growth pattern) cases.

  • Class III cases.

  • Cases with myofascial pain dysfunction (MPD) and myalgia.

  • Prior treatment involving temporomandibular joint surgery.

  • Cases with any skeletal disorders such as cherubism, osteoporosis.

Methods

Class II and Class I patients will be assessed using both ANB and Wits appraisal. The patient will then be diagnosed according to the diagnostic criteria for temporomandibular disorders (DC/TMD).9 Each participant will be asked for their written, informed consent. Based on the severity of TMD, skeletal Class II (vertical and horizontal growth pattern) cases will be graded using DC/TMD criteria. Both bilateral saggital and coronal MRI images will be obtained.

The articular disc of the TMJ on MRI will be identified and manually segmented with the help of technical expertise from the Faculty of Artificial Intelligence and Machine Learning (AIML) at FEAT College, Datta Meghe Institute of Higher Education and Research. After segmentation those images will be used as the dataset. The dataset showing the normal position of articular discs will be randomly split into a training data set and test set. To reduce the overfitting of the network, the dropout layer is placed behind the convolutional layers and max-pooling layers. Region of interests around the articular disc will be extracted from the datasets.

Region of interests will be automatically cropped from the images using Python 3.11 algorithms. Convolutional neural network (CNN) encoder-decoder named U-Net model architecture using the software Visual Studio Code with the Python programming language will be used to segment the articular disc on MRI. The variation in articular disc morphology will be evaluated and compared in Class II (vertical and horizontal growth pattern) with Class I Non-TMD cases using deep learning algorithms on MRI.

Outcomes

Primary outcome: To evaluate articular disc morphology in skeletal class II vertical and horizontal growth pattern.

Secondary outcome: To compare and correlate the variation in articular disc morphology with skeletal pattern.

Tertiary outcome: For doctors and radiologists, the ability to precisely recognise important features in MRI of the TMJ region will save time and increase accuracy. Also understanding the nature of variations between Class I and both types of Class IIs will help orthodontists to better predict the potential risk for development of TMDs and accordingly take precautions while carrying out treatment in such cases. Moreover, it can be used to automate TMD diagnosis and other smart applications. This is why automated segmentation of TMJ structures with MRI is clearly necessary.

Bias

Bias will be minimized by random selection of patients based on the inclusion and exclusion criteria.

Study sample

The calculation of sample size was carried out as follows:

Formula using mean difference

n1=n2=2Zα+Zβ2σ2δ2

Zα = 1.64

α = Type I error at 5% at both sides two tailed

Ζβ = 0.84 (1 - β) = power at 80%

σ = standard deviation

Primary variable: Articular disc variation (mm)

Class I TMD group = 1.38 ± 0.20 (John et al. (2020)10)

Class II (horizontal growth) TMD (mm)) group = 1.51 ± 0.20 (John et al. (2020)10)

Clinically relevant difference = 0.13

Pooled standard deviation = (0.20+0.20)/2 = 0.2

N1=21.64+0.8420.22/0.132=30

Total samples required = 30 per group.

Total sample size = 90

The sample would be divided into three groups based on the inclusion and exclusion criteria:

Group A (control group): 30 Class I (Non-TMD) cases.

Group B: 30 skeletal Class II (vertical growth pattern) cases.

Group C: 30 skeletal Class II (horizontal growth pattern) cases.

Statistical analyses

All the demographic outcome data will be presented using descriptive statistics for categorial variables in terms of frequency and percentage for continuous variables in terms of mean, standard deviation and median. Results will be analysed using SPSS version 27 (RRID:SCR_019096) for statistical analysis. Outcome variables will be tested for normality using the Kolmogorov-Smirnov test for continuous data.

All the samples will be distributed amongst the category of Class I non-TMD, Class II (vertical growth pattern) TMD cases and Class II (horizontal growth pattern) TMD cases as per the cephalometric measurements. Articular disc variation at different positions will be evaluated between these three groups using ANOVA or Kruskal Wallis test.

ANOVA or a Kruskal Wallis test will be used to find the result amongst the three groups for the outcome variable for the significance in mean difference. If the data are normally distributed an ANOVA test will be used and if the data are not normally distributed a non-parametric test (Kruskal Wallis) will be used to find the significant difference.

Dissemination

The original research will be published in a research article of an index journal. To ensure the original study receives the most exposure possible, the trial results will be released as open access.

Study status

The deep learning model is currently being prepared and the study will commence in May 2023.

Discussion

Nowadays, a lot of people have TMD due to the increased stress brought on by our fast-paced society. TMD is regarded as a collection of orofacial joint and muscle problems marked by pain, abnormal joint noises, and uneven jaw function. Although only 30% of subjects may be aware of such symptoms, the majority of research show that at least 50% of people have at least one symptom (such as muscle tenderness or joint clicking).11

There is variation in the morphology of the hard and soft tissue structures of TMJ specially of the articular disc. At times visualization and clarity of joint morphology and the articular disc may hinder the accuracy of diagnosis. What is the extent of variation and what type of disc morphology may cause a higher potential risk of TMD needs to be explored for early and more accurate prediction.

Hirata et al. (2007) studied the articular disc shape and location in patients with disc displacement, as well as the shape of the temporomandibular joint’s articular eminence. In the study, 14 individuals with bilateral disc displacement and no unilateral reduction were included. For evaluation, they employed magnetic resonance scans and showed that the chance of non-reducing disc displacement may be influenced by the shape of the articular disc and eminence.12

In 2020, John et al., observed that a Class II condition with a vertical growth pattern had the highest probability of developing a TMD with internal disk derangements and reduced anterior and posterior joint spaces compared to Class II with a horizontal growth pattern or Class I condition.10

Ito et al. (2022), conducted a study that assessed fully segmentation of the temporomandibular joint’s articular disc by automated means. Ten patients with anterior disc displacement and ten healthy control participants with normal articular discs were enrolled in the study. On MRI, they applied a semantic segmentation method based on deep learning. The study showed that this method for segmenting articular discs using automated deep learning on MRI generated encouraging preliminary results suggesting that the approach might be employed in clinical practice for the evaluation of temporomandibular disorders.3

Based on the findings of these previous studies, this planned study would enable us to recognize changes in articular disc shape, changes in articular disc location, and morphological abnormalities that can be evaluated by the deep learning network. The outcome of the treatment and diagnostic efficiency for TMD’s may improve using this technique.

Ethical considerations

Ethical approval has been granted by the Institutional Review Board of Datta Meghe Institute of Higher Education and Research, Sawangi, Wardha. (Reference number: DMIHER (DU)/IEC/2023/570 on 06/02/2023).

Comments on this article Comments (0)

Version 2
VERSION 2 PUBLISHED 20 Jul 2023
Comment
Author details Author details
Competing interests
Grant information
Copyright
Download
 
Export To
metrics
Views Downloads
F1000Research - -
PubMed Central
Data from PMC are received and updated monthly.
- -
Citations
CITE
how to cite this article
Surendran A, Shrivastav S and Srivastav G. Comparative evaluation and correlation of variations in articular disc morphology as assessed by automated segmentation using deep learning on magnetic resonance imaging (MRI) images in Class II (vertical) TMD cases, Class II (horizontal) TMD cases and Class I non-TMD cases [version 2; peer review: 1 approved]. F1000Research 2024, 12:855 (https://doi.org/10.12688/f1000research.133328.2)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Open Peer Review

Current Reviewer Status: ?
Key to Reviewer Statuses VIEW
ApprovedThe paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approvedFundamental flaws in the paper seriously undermine the findings and conclusions
Version 2
VERSION 2
PUBLISHED 15 Jul 2024
Revised
Views
3
Cite
Reviewer Report 22 Jul 2024
Martina Ferrillo, University of Catanzaro, Catanzaro, Italy 
Approved
VIEWS 3
The authors modified the paper according to my suggestions. ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Ferrillo M. Reviewer Report For: Comparative evaluation and correlation of variations in articular disc morphology as assessed by automated segmentation using deep learning on magnetic resonance imaging (MRI) images in Class II (vertical) TMD cases, Class II (horizontal) TMD cases and Class I non-TMD cases [version 2; peer review: 1 approved]. F1000Research 2024, 12:855 (https://doi.org/10.5256/f1000research.168904.r303157)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
Version 1
VERSION 1
PUBLISHED 20 Jul 2023
Views
23
Cite
Reviewer Report 12 Feb 2024
Martina Ferrillo, University of Catanzaro, Catanzaro, Italy 
Approved with Reservations
VIEWS 23
Temporomandibular disorders (TMD) are a collective of various symptoms caused by abnormalities in the temporomandibular joint (TMJ) and related structures. It affects 20% to 40% of the general population. Lifestyle habits, like alcohol consumption and smoking, are considered to be ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Ferrillo M. Reviewer Report For: Comparative evaluation and correlation of variations in articular disc morphology as assessed by automated segmentation using deep learning on magnetic resonance imaging (MRI) images in Class II (vertical) TMD cases, Class II (horizontal) TMD cases and Class I non-TMD cases [version 2; peer review: 1 approved]. F1000Research 2024, 12:855 (https://doi.org/10.5256/f1000research.146307.r232583)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
  • Author Response 15 Jul 2024
    Aathira Surendran, Department of Orthodontics, Sharad Pawar Dental College, Datta Meghe Institute of Higher Education and Research, Wardha, 442001, India
    15 Jul 2024
    Author Response
    Thank you ma'am for your thorough analysis of my article. I have made the following changes based on your review.

    1. Classification. Please refer to the Diagnostic Criteria for TMD ... Continue reading
COMMENTS ON THIS REPORT
  • Author Response 15 Jul 2024
    Aathira Surendran, Department of Orthodontics, Sharad Pawar Dental College, Datta Meghe Institute of Higher Education and Research, Wardha, 442001, India
    15 Jul 2024
    Author Response
    Thank you ma'am for your thorough analysis of my article. I have made the following changes based on your review.

    1. Classification. Please refer to the Diagnostic Criteria for TMD ... Continue reading

Comments on this article Comments (0)

Version 2
VERSION 2 PUBLISHED 20 Jul 2023
Comment
Alongside their report, reviewers assign a status to the article:
Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions
Sign In
If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

The email address should be the one you originally registered with F1000.

Email address not valid, please try again

You registered with F1000 via Google, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Google account password, please click here.

You registered with F1000 via Facebook, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Facebook account password, please click here.

Code not correct, please try again
Email us for further assistance.
Server error, please try again.