Interpretability And Reliability In Neural Network-Based Paediatric Thyroid Nodule Diagnosis: A Framework For Clinical Integration
Keywords:
Paediatric Thyroid Diagnosis, Neural Network Interpretability, AI in Paediatric Diagnostics, Multi-Pathway Attention Mechanism, Explainable AI (XAI), Clinical Validation, AI in Medical ImagingAbstract
This research introduces TI-PedThyroNet (Transparent and Interpretable Paediatric Thyroid Network), a novel framework enhancing interpretability and reliability in neural network-based thyroid nodule diagnosis specifically for paediatric populations. By integrating complementary interpretability techniques with uncertainty quantification, the methodology addresses the critical trust gap in AI-driven paediatric diagnostics where radiation exposure concerns and long-term implications of interventions require particular attention. Our multi-pathway attention mechanism optimizes feature extraction while providing granular explanations. Clinical validation with paediatric radiologists demonstrates significant improvements in diagnostic confidence (31% increase), decision-making speed (34% reduction in interpretation time), and trust metrics. TI-PedThyroNet achieves state-of-the-art performance (accuracy: 92.8%, sensitivity: 94.3%, specificity: 91.6%) while providing human-interpretable explanations and reliable uncertainty estimates, demonstrating considerable potential for clinical integration in paediatric settings.
Downloads
Metrics
References
S. Gupta et al., "Paediatric thyroid cancer: An update," J. Pediatr. Endocrinol. Metab., vol. 33, no. 5, pp. 585–599, 2020.
H. M. van Santen et al., "Paediatric differentiated thyroid carcinoma: clinical course and long-term follow-up," Endocr. Rev., vol. 42, no. 2, pp. 218–242, 2021.
J. D. Francis et al., "A comparison of thyroid ultrasound findings in children and adults: which nodular features are more worrisome in children?," AJR Am. J. Roentgenol., vol. 214, no. 6, pp. 1421–1425, 2020.
N. Norlen et al., "Risk factors for malignancy in paediatric thyroid nodules: A systematic review," Paediatrics, vol. 145, no. 4, p. e20192019, 2020.
M. Chen et al., "Long-term quality of life in adult survivors of paediatric differentiated thyroid carcinoma," J. Clin. Endocrinol. Metab., vol. 105, no. 7, pp. e2435–e2444, 2020.
S. M. Lauritsen et al., "Explainable artificial intelligence model to predict acute critical illness from electronic health records," Nat. Commun., vol. 11, no. 1, pp. 1–11, 2020.
K. Zhang et al., "Thyroid nodule classification using ultrasound image-guided vision transformer with attention mechanism," IEEE Trans. Med. Imaging, vol. 41, no. 11, pp. 3139–3150, 2022.
A. Amini et al., "Deep evidential regression," in Proc. Adv. Neural Inf. Process. Syst., 2020, pp. 14927–14937.
E. Beede et al., "A human-centered evaluation of a deep learning system deployed in clinics for the detection of diabetic retinopathy," in Proc. CHI Conf. Hum. Factors Comput. Syst., 2020, pp. 1–12.
L. Oakden-Rayner et al., "Hidden stratification causes clinically meaningful failures in machine learning for medical imaging," in Proc. ACM Conf. Health Inference Learn., 2020, pp. 151–159.
F. Arcadu et al., "Deep learning algorithm predicts diabetic retinopathy progression in individual patients," NPJ Digit. Med., vol. 2, no. 1, pp. 1–9, 2019.
A. Dosovitskiy et al., "An image is worth 16x16 words: Transformers for image recognition at scale," in Proc. Int. Conf. Learn. Represent., 2021.
J. Kim et al., "Differentiation of solid thyroid nodules using an artificial intelligence algorithm based on ultrasound data," Sci. Rep., vol. 10, no. 1, pp. 1–8, 2020.
J. Lei et al., "Explaining the black-box model: A survey of local interpretation methods for deep neural networks," Neurocomputing, vol. 440, pp. 76–93, 2021.
F. Yang et al., "A deep learning approach for thyroid ultrasound image segmentation," Med. Phys., vol. 47, no. 4, pp. 1834–1845, 2020.
J. Y. Park et al., "Ensemble-based deep learning model for thyroid nodule diagnosis in ultrasound imaging with interpretability," Diagnostics, vol. 11, no. 6, p. 987, 2021.
A. Mitani et al., "Deep learning-based image analysis for paediatric thyroid nodule diagnosis," J. Pediat. Endocrinol. Metab., vol. 33, no. 8, pp. 1057–1065, 2020.
L. G. Vergara et al., "Explainable artificial intelligence for paediatric thyroid ultrasound: A pilot study," Pediatr. Radiol., vol. 51, no. 6, pp. 941–950, 2021.
F. Borson et al., "Interpretable machine learning models for medical domain: A perspective review from paediatric radiology," J. Imaging, vol. 7, no. 10, p. 212, 2021.
S. Goutte et al., "The genetics of paediatric thyroid nodules and differentiated thyroid cancer," Genes, vol. 12, no. 8, p. 1158, 2021.
W. Liu et al., "Current insights into the molecular genetics of paediatric thyroid carcinoma," Front. Endocrinol., vol. 12, p. 722289, 2021.
A. K. Lam et al., "Comparison of histopathological features and expression of selected genes in follicular thyroid tumors and their counterparts in paediatric age," Cancers, vol. 13, no. 17, p. 4336, 2021.
M. Nishino et al., "Ultrasound-guided fine-needle aspiration cytology of paediatric thyroid nodules: A multi-institutional review of 324 cases," Cancer Cytopathol., vol. 129, no. 5, pp. 389–397, 2021.
D. Cheng et al., "Deep learning methods for paediatric medical imaging: Current state and future opportunities," Artificial Intelligence in Medicine, vol. 122, p. 102207, 2021.
Y. Zhou et al., "Hybrid models for segmentation of paediatric thyroid nodules in ultrasound images," IEEE Trans. Biomed. Eng., vol. 68, no. 11, pp. 3372–3383, 2021.
S. Lin et al., "Federated learning for medical imaging in paediatric applications: A systematic review," J. Am. Med. Inform. Assoc., vol. 28, no. 12, pp. 2691–2704, 2021.
Downloads
Published
How to Cite
Issue
Section
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
You are free to:
- Share — copy and redistribute the material in any medium or format
- Adapt — remix, transform, and build upon the material for any purpose, even commercially.
Terms:
- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
- No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.
