Thesis title: ARTIFICIAL INTELLIGENCE IN PREDICTING TREATMENT AND FOLLOW UP OF THYROID DISORDERS COMPARED WITH FINE NEEDLE ASPIRATION
Abstract
Introduction: Thyroid nodules categorized as Bethesda III and IV remain
diagnostically challenging due to indeterminate cytology, often leading to
unnecessary surgeries or delayed interventions. Fine Needle Aspiration Biopsy
(FNAB) indeterminate in these cases, prompting the need for alternative diagnostic
tools.
Aim: To evaluate the diagnostic accuracy and clinical utility of an Artificial
Intelligence (AI) classification system for thyroid nodules with indeterminate
cytology (Bethesda III and IV), and compare its performance with histological
outcomes and clinical decisions.
Methods: This prospective study was conducted at the Service of Endocrinology
and Surgery, University Hospital Center “Mother Teresa” in Tirana and Memorial
Hospital in Fier, Albania, between January 2022 and June 2024. A total of 138
patients with Bethesda III and IV nodules were included. An AI diagnostic model
classified each nodule as benign or malignant based on imaging and cytological
features. These classifications were compared with final histological diagnoses and
corresponding treatment decisions.
Results: Across all 138 patients, the AI classifier showed balanced accuracy
(~87%) with strong rule-out performance (NPV ~90%). FNAB was excellent where
evaluable (II/V/VI)—near-perfect specificity—but doesn’t cover indeterminate
nodules. In the Bethesda III–IV subgroup (n=53), AI maintained solid accuracy
(~87%), with high NPV (95%) and moderate PPV (62%). Management aligned
with AI labels: AI-benign was usually observed (60%), AI-malignant usually
operated (85%), with a significant association (p=0.009; OR≈8). In multivariable
analysis, an AI-malignant label independently predicted surgery (OR 7.06, 95% CI
1.18–42.42; p=0.032) after adjusting for age, sex, laterality, and nodule structure.
Mixed vs cystic structure was associated with lower odds of surgery (OR 0.18, 95%
CI 0.03–0.97; p=0.046).
Conclusions: AI models offer a accurate, non-invasive solution for managing
indeterminate thyroid nodules. Their ability to classify all cases—especially where
10
FNAB fails—places AI as a transformative tool in thyroid cancer diagnostics and
surgical decision-making.