Artificial intelligence is often researched and used to reduce subjectivity in enforcing several diseases, including thyroid nodules. Until now, the diagnosis of thyroid nodules still relies on  Ultrasonograph examination (USG). If thyroid ultrasonography results suggest a malignant lesion, perform a more invasive evaluation (fine needle aspiration biopsy (FNAB).

Artificial Intelligence in the Diagnosis of Thyroid Nodules



Thus, the diagnosis of thyroid nodules is a challenge for clinicians because the subjectivity of thyroid ultrasonography and the fine-needle aspiration biopsy (FNAB) are often inconclusive. It is necessary to collect repeated specimens for histopathological examination.

This inclusiveness makes evaluating thyroid nodules ineffective, inefficient, and expensive. Cheng et al. developed an ultrasound-based diagnostic algorithm called the Thyroid Imaging Report Data System (TI-RADS). Furthermore, methoxy-isobutyl-isonitriles-single emission computed tomography (MIBI-SPECT) is also used for the diagnosis of thyroid nodules. However, the diagnosis of thyroid nodules is controversial because of its complexity.

Trying to solve the problem, the researchers examined the accuracy of a combination of artificial intelligence thyroid (AI thyroid) and imaging (such as magnetic resonance imaging (MRI), X-rays, and ultrasonography). Combining these modalities is considered to help diagnose thyroid nodules and is thought to be a way out of it.



Artificial Intelligence Thyroid (AI Thyroid) in the Diagnosis of Thyroid Nodules

Previous studies classified thyroid nodules by handcrafted-based and deep-learning-based methods. The handcrafted-based method uses image extraction, which is further analyzed with a support vector machine (SVM) so that the characteristics of the thyroid nodule are examined.

The deep-learning-based method is used to overcome the handcrafted-based method's disadvantage, namely the tool's dependence on images of thyroid nodules. In this method, a convolutional neural network (CNN) tool analyzes thyroid nodule ultrasonography.



Accuracy of Artificial Intelligence in the Diagnosis of Thyroid Nodules

How accurate is artificial intelligence (AI) for diagnosing thyroid nodules? Artificial Intelligence is expected to reduce the thyroid biopsy's need if AI intelligence is equivalent to current conventional diagnostic methods.

In their study, Wang et al. attempted to develop a deep-learning method based on an algorithm in the form of a YOLOv2 neural network. In that study, they reported that thyroid AI could precisely diagnose areas of lesions with a higher value than radiological findings.

Also, there are reports that Artificial Intelligence has a sensitivity of 90.5%, a positive predictive value of 95.2%, a negative predictive value of 80.99%, and 90.31% accuracy. These percentages were not significantly different from the radiological findings. Additionally, Artificial Intelligence has a higher specificity than radiological findings (89.91% versus 77.98%).

Compared with thyroid ultrasonography with experienced operators, Artificial Intelligence has the same sensitivity and accuracy for diagnosing malignant thyroid nodules, even better diagnosing benign thyroid nodules.

Recently, Thomas and Haertling reported that deep-neural-based AI had sensitivity, specificity, accuracy, and negative predictive values ​​of 87.8%, 78.5%, 65.9%, and 93.2%.

Compared with USG, artificial intelligence has better sensitivity, specificity, and positive predictive value in stratifying thyroid cancer risk. These advantages are breakthrough, considering that previously thyroid diagnosis based on artificial intelligence was limited to malignant lesions or thyroid carcinoma.

Yu et al. conducted a similar study and reported that:
  • AI has 90% accuracy, 88.24% sensitivity, and 90.91% specificity
  • deep-learning-based on 86% accuracy, 81.25% sensitivity, 90.91% specificity.
These reports indicate that AI is better than conventional methods and equivalent to each other.

In detecting malignant lesions, It should be noted that most of the artificial intelligence developed today is based on the discovery of thyroid papillary carcinoma lesions. Thus, further research is needed for other types of carcinoma, such as lymphoma or medullary carcinoma. It should also be noted that there is a selection bias in studies assessing the role of artificial intelligence in diagnosing thyroid nodules, so further research is still needed.



Recommendations regarding the Diagnosis of Thyroid Nodules

The American Thyroid Association recommends performing thyroid ultrasonography to evaluate cervical lymph nodes in all patients with suspected thyroid nodules. The pattern on the thyroid ultrasonography findings and the nodule's characteristics will assist in the decision-making for FNAB.

FNAB is still considered the most accurate modality for evaluating thyroid nodules, especially for nodules ≥1 cm.



Summary
Until recently, the diagnosis of thyroid nodules had generally depended on ultrasonography and FNAB examinations. Unfortunately, both modalities involve subjective judgment and operator skill. Additionally, the FNAB examination often causes discomfort to patients because it usually takes more than one specimen collection for this examination.

These problems make the diagnosis of thyroid nodules is ineffective and inefficient. Therefore, the use and development of an artificial intelligence-based algorithm system aim to help diagnose thyroid nodules and reduce the need for FNAB.

Currently, AI technology for diagnosing thyroid nodules consists of two methods: handcrafted-based and deep learning-based. Several existing studies report that the two methods are equivalent to each other and superior to conventional diagnosis.

However, studies are limited to detecting papillary thyroid carcinoma, whereas its accuracy and role in detecting medullary carcinoma and lymphoma are not definitive. By looking at current developments, artificial intelligence could play an even more role in diagnosing thyroid nodules in the future. Therefore, studies with larger samples and better methods are needed to support thyroid nodules' diagnostic development.


References
1. Ali S, Siperstein A, Sadow P, Golding A, Kennedy G, Kloos R, et al. Extending expressed RNA genomics from surgical decision making for cytologically indeterminate thyroid nodules to targeting therapies for metastatic thyroid cancer. Cancer Cytopathology. 2019;127(6):362-9.
2. Chang C, Chen S, Tsai M. Application of support-vector-machine-based method for feature selection and classification of thyroid nodules in ultrasound images. Pattern Recognition. 2010;43(3494-3506).
3. Haugen BR, Alexander EK, Bible KC, Doherty GM, Mandel SJ, Nikiforov YE, et al. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid. 2015;26(1):1-133.
4. Horvath E, Majlis S, Rossi R, Franco C, Niedman JP, Castro A, et al. An Ultrasonogram Reporting System for Thyroid Nodules Stratifying Cancer Risk for Clinical Management. Journal of Clinical Endocrinology and Metabolism. 2009;94(5):1748-51.
5. Listewnik MH, Birkenfeld B, Piwowarska-Bilska H, Cichon-Bankowska K, Iglinska-Wagner L, Watrak W, et al. The Application of SPECT/CT Scintigraphy With MIBI-Tc⁹⁹(m) in the Diagnosis of Thyroid Nodules - A Preliminary Report. Endocrinology Poland. 2010;61(5):422-6.
6. Nguyen DT, Pham TD, Batchuluun G, Yoon HS, Ryoung PK. Artificial Intelligence-Based Thyroid Nodule Classification Using Information from Spatial and Frequency Domains. Journal of Clinical Medicine. 2019;8(11):1976.
7. Sundar K, Rajamani K, Sai S. Exploring image classification of thyroid ultrasound images using deep learning. Proceedings of the International Conference on ISMAC in Computational Vision and Bio-Engineering. Cham: Springer. 2018. p. 1635-41.
8. Thomas J, Haertling T. AIBx, artificial intelligence model to risk stratify thyroid nodules. Thyroid. 2020;30(6):1.
9. Wang L, Yang S, Yang S, Zhao C, Tian G, Gao Y, et al. Automatic thyroid nodule recognition and diagnosis in ultrasound imaging with the YOLOv2 neural network. World Journal of Surgical Oncology. 2019;2019(17):12.
10. Yu Q, Jiang T, Zhou A, Zhang L, Zhang C, Xu P. Computer-aided diagnosis of malignant or benign thyroid nodes based on ultrasound images. European Archives of Otorhinolaryngology. 2017;274:2891-7.