Journal of Oral Health Sciences and Dentistry
Research Article Volume: 1 & Issue: 2
Journal Menu
- Home
- Aims and Scope
- Author Guidelines
- Editorial Board
- Reviewer Board
- Current Issue
- Archive
- Journal Contact
Indexed & Abstracted In
Abstract
Artificial intelligence (AI) has revolutionized cephalometric analysis in orthodontics, enabling automated landmark detection and diagnostic measurements. However, the vulnerability of these systems to adversarial machine learning attacks remains unexplored, posing potential risks to patient safety and clinical decision-making. This study investigates the susceptibility of AI-based cephalometric analysis systems to adversarial attacks and evaluates their impact on diagnostic accuracy and clinical outcomes. We developed a deep learning model for automated cephalometric landmark detection using a dataset of 1,200 lateral cephalometric radiographs. We then generated adversarial examples using four attack methods: Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), Carlini & Wagner (C&W), and Deep Fool. The impact on landmark detection accuracy, angular measurements, and treatment planning decisions was systematically evaluated. Adversarial perturbations imperceptible to human observers (average ε = 0.03) caused significant degradation in model performance. FGSM attacks reduced landmark detection accuracy from 94.2% to 61.7%, while more sophisticated PGD attacks decreased accuracy to 43.8%. Critical angular measurements showed errors up to 8.4°, potentially leading to incorrect treatment classifications in 37% of cases.
Keywords: Adversarial machine learning, cephalometric analysis, medical imaging security, deep learning robustness, orthodontic AI, healthcare cybersecurity.
AUTHOR TESTIMONIALS
Trusted by Researchers
Experiences from authors who have published with SCIVOLVE.
I had a very positive experience working with the Journal of Case Reports and Reviews in Medicine. The publication process was quick and well organized, every step was clearly explained, and the editorial team was always responsive and helpful.
I really enjoyed communicating with the editorial staff. The editorial team provided clear and efficient communication throughout the process. The manuscript submission process was straightforward, and the editorial team made it easy to address the necessary corrections. I appreciated the professional, patient, and supportive approach of the editorial staff, which has encouraged me to consider publishing my future work.
I had an excellent experience publishing my manuscript with the journal. The communication with the editorial team was clear, professional, and efficient throughout the entire publication process. I truly appreciated their support and responsiveness.
As I pursue a Doctor of Juridical Science degree, I am pleased to have worked with Scivolve Publishers on publishing my article on allergies. In my work to advance women’s human rights and help humanity, it is vital to have publishing opportunities that respect research integrity. From the first contact with the Editorial Manager to the final stage, the process was precise, transparent, and exemplary. Nothing was overpromised, and the handling of sources and references was also excellent. I warmly recommend Scivolve Publishers for their integrity, professionalism, and the respect with which they conduct their work.
From submission to final publication, the process was organized and easy to follow. The editorial communication made the overall experience straightforward and professional.