Pathological diagnosis of thyroid nodules directly from ultrasonography by a weakly supervised deep learning framework

By
Xiao-Wen Hou
Mei-Ling He
Meng-Yue Ye
Jian-Xin Ji
Liu-Ying Wang
Wei Zhang
Yue Zhao
Zhen-Zhao Sun
Hui-Ru Jiang
Ping Li
Ji-Hong Wang
Fan-Chao Shi
Shu-Xin Sun
Lei Cao
June 1, 2026
0 min

Frontiers In Endocrinology

Objective:

To develop a dual attention-guided deep learning framework for inferring histological status directly from thyroid ultrasonography without expensive image-level annotations.

Key Findings:

ThyUS2Path achieved AUCs of 0.754 ± 0.035 and 0.735 ± 0.029 in cross-validation.
The algorithm outperformed Maxpool and Meanpool methods significantly.
Good prediction performance was observed in the external test set with AUROCs of 0.70~0.80 and AUPRCs of 0.78~0.83.

Interpretation:

The approach provides a feasible method to link ultrasound phenotypes with histological reports, potentially enhancing non-invasive thyroid cancer diagnosis.

Limitations:

The study is retrospective and may be limited by the datasets used.
The performance may vary with different populations or imaging conditions.

Conclusion:

ThyUS2Path demonstrates potential in augmenting clinicians' capabilities in thyroid cancer diagnosis through non-invasive methods.

Pathological diagnosis of thyroid nodules directly from ultrasonography by a weakly supervised deep learning framework

Objective:

Key Findings:

Interpretation:

Limitations:

Conclusion:

Original Source(s)

Pathological diagnosis of thyroid nodules directly from ultrasonography by a weakly supervised deep learning framework

Related Content

ISG15 deficiency in hepatic stellate cells promotes TGFβ2-induced liver fibrosis by counteracting CREB1 ISGylation

Editorial: Technological evolution in minimally-invasive radical prostatectomy. From laparoscopic to single-port robotically assisted

Antibody-Drug Conjugates for Locally Advanced and Metastatic Urothelial Carcinoma: A Systematic Review and