University of Illinois at Chicago
Browse
RHEE-PRIMARY-2024.pdf (2.77 MB)

Evaluation of Anterior Open Bite Treatment Strategies and Outcomes Using Artificial Intelligence

Download (2.77 MB)
thesis
posted on 2024-05-01, 00:00 authored by Matthew Aaron Tjernlund Rhee
Objective: The creation of a data set used for supervised machine learning algorithms to accurately classify patients based on their initial presentation. Examine if patients with anterior open bites, in which extraction treatment leads to successful treatment, can be grouped/clustered based on their initial presentation. Methods: This was a retrospective chart review study. Participants selected were treated (started and finished) at the University of Illinois at Chicago Orthodontic Clinic (UIC). These patients presented with anterior open bites larger than or equal to 2mm and were treated successfully with or without extraction of one or multiple teeth. Successful treatment was defined as achieving a positive overbite measurement. Pre-treatment clinical characteristics collected included age, race, sex, molar classification, and dental crowding. Radiographic data collected includes cephalometric measurements found in the “Rickett’s Advanced” and “ABO 2012” analysis. Results: 1100 Patients were identified with anterior open bites greater than or equal to 2mm that had initiated and finished treatment at UIC. After the remaining inclusion and exclusion criteria were applied the remining sample size was 115 patients. Of these patients with successful treatment, 83 were female (72.2%) and 32 (27.8%) were male. Further divergence into extraction and non-extraction groups showed similar percentages. In the extraction group of 55 subjects, 40 (72.7%) were female and 15 (27.3%) were male. In the non-extraction group of 60 subjects, 43 (71.7%) were female and 17 (28.3%) were male. The ages of the total group ranged from 10 to 51 with an overall average of 17 years. The ages of the extraction group ranged from 11-38 with an average age of 15.6 years. The ages of the non-extraction group ranged from 10-51 with an average age of 18.3 years. After final data cleaning and preliminary characterization of the dataset, different machine learning algorithms were run with the data. Initial results from the Random Forest algorithm has shown that machine learning can predict whether extraction should occur 82.74% of the time. Additional data supporting which characteristics were strong predictors of the decision will be sought with subsequent iterations of the algorithms. Conclusions: With the initial results it appears that machine learning algorithms can predict extraction treatment with an accuracy of 82.74%. Characteristics most affecting this result include crowding, SNA, SN-Palatal Plane, Facial Axis, Total Facial Height, Lower Facial Height, and FMA. With a success prediction at this frequency, it is imperative that current machine learning algorithms be used judiciously; treatment considerations utilizing this technology must also integrate a very strong base in the science and practice of orthodontia and should absolutely not be used as the sole tool used to make treatment decisions.

History

Advisor

Flavio Sanchez

Department

Orthodontics

Degree Grantor

University of Illinois Chicago

Degree Level

  • Masters

Degree name

MS, Master of Science

Committee Member

Mohammed Elnagar Veerasathpurush Allareddy Ahmet Cetin

Thesis type

application/pdf

Language

  • en

Usage metrics

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC