Thesis
Evaluation of Performance of Deep Learning Algorithms in Detecting and Diagnosing Dental Carious Lesions Using Intraoral Radiographic Imaging: A Systematic Review and Meta-Analysis
University of Iowa
Master of Science (MS), University of Iowa
Summer 2022
DOI: 10.25820/etd.006476
Abstract
IntroductionDeep learning, a form of artificial intelligence is gaining increasing interest and has the potential to aid clinicians in diagnostic imaging. There is a tremendous increase in research and development of various deep learning algorithms in recent years. In the past 4-5 years many studies have been conducted to test the performance of these new machine learning approaches. However, there are limited studies that compare the performance of the algorithm with experts, especially in dentistry. The reliability and robustness of the recent data remain to be explored. With the rapid developments in this field, there is a need for critical analysis of these studies as there are always concerns regarding study design, data collection, and unbiased comparison of the results. Systematic reviews and meta-analyses are effective methods to determine the consistency of the study outcomes and will also guide future research.
Methods
This study evaluated the accuracy of deep learning algorithms in identifying dental caries on intraoral radiographs. For this project, we searched MEDLINE, Embase, Web of Science, and CINAHL databases to identify clinical studies which used deep learning algorithms to detect caries. Studies performed on extracted teeth and simulated caries lesions were excluded. These studies were critically reviewed, and a formal qualitative assessment was performed. Studies comparing the performance of the algorithms with dentists were further selected for meta-analysis. Diagnostic performance data e.g., sensitivity, specificity, accuracy, etc. were extracted and analyzed.
Results
Our search identified 886 studies, of which 21 were included in this systematic review. Heterogeneity in the study design and inconsistencies in the methodology were observed. For example, only 21% of the studies used the appropriate method to determine the ground truth. Owing to the heterogeneity among these studies random-effects model was used for qualitative assessment. Only 7 studies compared the performance of the artificial intelligence algorithm with dental experts. The accuracy of the algorithms was high and ranged from 74% to 97%. The sensitivity ranged from 20% to 86.6%.
Conclusions
The accuracy and sensitivities of the deep learning algorithms are slightly higher than that of the dental experts. However, the specificity and precision are lower than those of dental experts. Given the methodology and reporting inconsistencies, these findings cannot be generalized and applied clinically. Standardization in the methodology and transparency in the reporting of the studies is required.
Details
- Title: Subtitle
- Evaluation of Performance of Deep Learning Algorithms in Detecting and Diagnosing Dental Carious Lesions Using Intraoral Radiographic Imaging: A Systematic Review and Meta-Analysis
- Creators
- Manila Shindé
- Contributors
- Trishul Allareddy (Advisor)Xian Jin Xie (Committee Member)Shankar Rengasamy Venugopalan (Committee Member)Sindhura Anamali (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Oral Science
- Date degree season
- Summer 2022
- DOI
- 10.25820/etd.006476
- Publisher
- University of Iowa
- Number of pages
- ix, 80 pages
- Copyright
- Copyright 2022 Manila Shindé
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 55-65).
- Public Abstract (ETD)
Artificial intelligence is becoming popular and has been applied in the field of radiographic imaging. There is a tremendous increase in research and development of various deep learning algorithms in recent years. In the past 4-5 years many research studies have been published which are based on the use of artificial intelligence to caries on dental radiographs. However, there are limited studies that compare the performance of the algorithm with dental experts. The results of these studies are variable and hence there is a need to compare these studies. We need to explore the reliability and strength of these studies. Unbiased, critical analysis of these studies will help us understand the usefulness of artificial intelligence in detecting caries. This project utilizes the scientific method of “Systematic Reviews” to assess the validity of these studies and their results.
This study sought to evaluate whether artificial intelligence has accuracy comparable to dentists’ in identifying caries on dental radiographs. After reviewing 886 records from several databases, we identified 21 studies related to this topic. These studies were critically reviewed. It was observed that in several studies there was a lack of transparency in reporting the research methodology and results. Only 7 studies compared the performance of the artificial intelligence algorithm with dental experts. The accuracy of the algorithms was high and ranged from 74% to 97%. The sensitivity ranged from 20% to 86.6%. Overall inconsistencies in the study designs and reporting made it difficult to quantify and compare the performance of these algorithms with that of the dentists.
project demonstrated that although artificial intelligence can be used to identify caries on dental radiographs, the clinical applicability remains to be determined.
- Academic Unit
- Oral Pathology, Radiology and Medicine
- Record Identifier
- 9984285249002771
Metrics
65 Record Views