Dataset Information

Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population.

ABSTRACT: Artificial Intelligence (AI) models for medical diagnosis often face challenges of generalizability and fairness. We highlighted the algorithmic unfairness in a large thyroid ultrasound dataset with significant diagnostic performance disparities across subgroups linked causally to sample size imbalances. To address this, we introduced the Quasi-Pareto Improvement (QPI) approach and a deep learning implementation (QP-Net) combining multi-task learning and domain adaptation to improve model performance among disadvantaged subgroups without compromising overall population performance. On the thyroid ultrasound dataset, our method significantly mitigated the area under curve (AUC) disparity for three less-prevalent subgroups by 0.213, 0.112, and 0.173 while maintaining the AUC for dominant subgroups; we also further confirmed the generalizability of our approach on two public datasets: the ISIC2019 skin disease dataset and the CheXpert chest radiograph dataset. Here we show the QPI approach to be widely applicable in promoting AI for equitable healthcare outcomes.

SUBMITTER: Yao S

PROVIDER: S-EPMC10912763 | biostudies-literature | 2024 Mar

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population.

Yao Siqiong S Dai Fang F Sun Peng P Zhang Weituo W Qian Biyun B Lu Hui H

Nature communications 20240304 1

Artificial Intelligence (AI) models for medical diagnosis often face challenges of generalizability and fairness. We highlighted the algorithmic unfairness in a large thyroid ultrasound dataset with significant diagnostic performance disparities across subgroups linked causally to sample size imbalances. To address this, we introduced the Quasi-Pareto Improvement (QPI) approach and a deep learning implementation (QP-Net) combining multi-task learning and domain adaptation to improve model perfor ...[more]

PMID: 38438371

Similar Datasets

Project description:BackgroundSince its inception, the Bethesda System for Reporting Thyroid Cytopathology (TBS) has been widely adopted. Each category conveys a risk of malignancy and recommended next steps, though it is unclear if each category also predicts the type and extent of malignancy. If so, this would greatly expand the utility of the TBS by providing prognostic information in addition to baseline cancer risk.MethodsAll patients prospectively enrolled into the authors' thyroid nodule database from 1995 to 2013 with histologically proven malignancy were analyzed. The primary ultrasound-guided fine-needle aspiration cytology (AUS, atypia of unknown significance; FN, follicular neoplasm; SUSP, suspicious; M, malignant) was correlated with the type of thyroid cancer and histological features known to impact prognosis and recurrence, including lymph node metastasis (LNM), lymphovascular invasion, and extrathyroidal extension (ETE). Primary cytology was separately correlated with higher risk malignancy.ResultsA total of 1291 malignancies were identified, with primary cytology AUS in 130 cases, FN in 241 cases, SUSP in 411 cases, and M in 509 cases. AUS, SUSP, and M cytology were progressively associated with an increasing risk of high-risk disease (p < 0.001), LNM (p < 0.001), ETE (p < 0.001), and margin positivity (p < 0.001). Notably, 71% of malignancies with AUS cytology were follicular variants of papillary thyroid cancer compared with 63% with SUSP cytology and only 20% with M cytology. In contrast, high-risk malignancies were diagnosed in only 4% with AUS cytology, but 9% and 27% with SUSP and M cytology, respectively. FN conveyed a significantly increased risk of follicular thyroid carcinoma compared with all other types (28% vs. 2%; p < 0.001). A composite endpoint of recurrence, distant metastases, and death similarly increased as cytology progressed from AUS to SUSP to M (p < 0.001).ConclusionIn addition to predicting cancer prevalence, the TBS also imparts important prognostic information about cancer type, variant, and risk of recurrence. These data extend the utility of TBS classification by fostering an improved understanding of the risk posed by any confirmed malignancy.

Dataset Information

Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population.

Publications

Enhancing the fairness of AI prediction models by Quasi-Pareto improvement among heterogeneous thyroid nodule population.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets