Open access peer-reviewed chapter - ONLINE FIRST

Role of Artificial Intelligence in Thyroid Disorders

Written By

Qaviullah Mian and Nebiyou Bayleyegn

Submitted: 16 April 2025 Reviewed: 05 May 2025 Published: 09 July 2025

DOI: 10.5772/intechopen.1010909

Recent Advances in Thyroid Disorders IntechOpen
Recent Advances in Thyroid Disorders Edited by Ifigenia Kostoglou-Athanassiou

From the Edited Volume

Recent Advances in Thyroid Disorders [Working Title]

Dr. Ifigenia Kostoglou-Athanassiou and Dr. Panagiotis Athanassiou

Chapter metrics overview

1 Chapter Downloads

View Full Metrics

Abstract

Artificial intelligence (AI), in its simplest form, uses computers to interpret or solve a simple or complex task, making our lives easier and better. Machine learning and deep learning are two fields of AI in medicine. Neural networks that resemble the neurons of the human nervous system are being used to enhance the processing of text, images, and clinical records for improved diagnostic efficiency, achieving high accuracy. With recurrent neural networks (RNNs), thyroid samples can be classified as cancerous or goiter, enhancing diagnostic capabilities for pathologists. There is a need for improved models to better understand hypothyroidism, hyperthyroidism, and thyroid cancer and facilitate novel therapeutic developments. Computer vision aids in the precise identification of vital structures during open and Minimal Access Surgery, reducing morbidity and fatal complications. As part of personalized medicine, pre-operative imaging data are analyzed for adequate pre-surgical planning and intraoperative real-time guidance during robotic-assisted thyroid surgery. Medical education has touched new horizons by using AI to provide automatic evaluation metrics while training for surgical skills. Holograms are being used to dissect the different layers of the thyroid gland to understand the peculiar anatomy of each patient. The use of virtual reality has revolutionized undergraduate and postgraduate education, allowing several students to be mentored and assessed simultaneously.

Keywords

  • artificial intelligence
  • machine learning
  • deep learning
  • neural network
  • thyroid disorders
  • thyroid dysfunction
  • thyroid cancer
  • thyroidectomy
  • computer vision

1. Introduction

The thyroid gland plays a central role in maintaining homeostasis in the human body through its hormones, specifically T3 and T4. These hormones are involved in various functions, including, but not limited to, the regulation of metabolism, thermogenesis, and growth and development. Although anatomically impalpable in normal conditions, its dysfunction may result in the most pronounced imbalance in the human body, even leading to vital compromise [1]. The thyroid gland is regulated by a feedback mechanism completed by the hypothalamus and the pituitary gland, which release thyrotropin-releasing hormone (TRH) and thyroid-stimulating hormone (TSH).

Physicians in the modern era face certain undeniable challenges in patient care related to the thyroid gland. Subjective interpretations of imaging, ambiguous biopsy results, and delays in thyroid function test analysis are some of the difficulties that traditional diagnostic and treatment approaches for thyroid disorders, including hypothyroidism and hyperthyroidism, encounter. These restrictions may result in incorrect diagnosis, postponed start of treatment, and challenges in providing patients with the best possible, individualized care.

Since its inception in 1956, artificial intelligence (AI) in medicine has shown visible progress in the 2000s, when patient electronic medical records and other electronic resources were analyzed to develop evidence-based responses in medicine [2]. This, along with technological advances in computer hardware and software, enabled humans to utilize natural language processing (NLP), a branch of AI, to create computer interfaces that facilitate conversations ranging from superficial to meaningful communication; Eliza, Siri, and Alexa, to name a few. Pharmabot was developed to educate patients on medication related to pediatrics, and similarly, Mandy was made to assist in primary practice. Machine learning (ML) entails analyzing extensive datasets to identify unique patterns and generate predictions, often termed supervised learning. Other techniques employ the use of unlabeled data, which is called unsupervised learning, and reinforcement learning, where a “rewards/punishments” system is used to provide feedback to the machine to improve its algorithm, whereas deep learning (DL), an enhanced form of ML, comprises algorithms to develop artificial neural networks, mimicking some aspects of the human brain, allowing the computer to make decisions independently [3].

AI has the potential to revolutionize patient care by enhancing screening for diseases, patient-doctor encounters, diagnostic workups, treatment modalities, rehabilitation medicine, and healthcare delivery.

This chapter systematically reviews the evidence for AI applications in thyroid disorder management while maintaining a critical perspective on implementation challenges. The discussion is organized according to the clinical workflow: diagnosis, treatment planning, and outcome monitoring, with special attention to thyroid cancer as a distinct domain where AI shows particular promise. Ethical and legal challenges have been addressed. Lastly, the impact of computer vision and augmented reality in medical education at both undergraduate and postgraduate levels is expanded upon as well.

Advertisement

2. Current challenges in thyroid disorder diagnosis and management

Clinicians face significant diagnostic and management challenges of thyroid disorders, encompassing conditions such as hypothyroidism, hyperthyroidism, and thyroid cancers. Traditionally, diagnostic modalities and treatment strategies have certain limitations that can impede better patient outcomes.

This section dives deep into the challenges mentioned previously to emphasize the need for innovative solutions to enhance patient care.

2.1 Limitations of traditional diagnostic methods

2.1.1 Subjectivity in interpretation

Ultrasound imaging is a cornerstone in evaluating thyroid nodules. However, since it is operator-dependent, this inter-observer variation often compromises its efficacy. The interpretation of ultrasound images of pathology in one patient need not always coincide, which could lead to an inconsistent assessment of the characteristics of the thyroid nodules, and a malignancy might be overlooked. This subjectivity can result in either overdiagnosis or underdiagnosis, affecting subsequent clinical decisions [4].

2.1.1.1 False positives and negatives in fine-needle aspiration biopsies

Fine-needle aspiration (FNA) biopsy is a standard procedure for evaluating thyroid nodules. Despite its widespread use, FNA has a notable rate of inconclusive results, ranging from 15% to 30%. Such indeterminate findings often necessitate repeat biopsies or additional diagnostic procedures, leading to patient anxiety, increased healthcare costs, and potential delays in definitive diagnosis [5].

2.1.1.2 Delayed diagnosis due to manual analysis of thyroid function tests

Thyroid function tests (TFTs), including measurements of serum thyroid-stimulating hormone (TSH) and thyroid hormones (T3 and T4), are essential for diagnosing thyroid dysfunctions. However, the manual interpretation of these tests can be time-consuming and is susceptible to human error. Delays in analyzing TFTs can postpone the initiation of appropriate treatment, potentially exacerbating patient conditions [6].

2.2 Challenges in treatment planning and monitoring

2.2.1 Personalized treatment needs

The management of hypothyroidism typically involves levothyroxine replacement therapy. Determining the optimal dosage is complex and influenced by patient age, weight, residual thyroid function, and comorbidities. Standard dosing guidelines recommend 1.6–1.8 μg/kg for overt hypothyroidism and 0.8–1 μg/kg for subclinical cases. However, individual responses vary, necessitating frequent monitoring and dose adjustments to achieve euthyroidism [7].

2.2.2 Monitoring disease progression

Post-treatment surveillance, especially in thyroid cancer patients, is critical for early detection of recurrence. Traditional monitoring methods, including periodic imaging and serum thyroglobulin measurements, may not always detect recurrence promptly. Deep learning models have been developed that analyze comprehensive data on the clinicopathological features of differentiated thyroid cancer to monitor disease progression and recurrence with high accuracy. This approach has enabled personalized treatment, allowing the surgeon to intervene promptly, improving patient care and outcomes significantly in thyroid cancer management [8].

2.3 Need for innovative solutions

The limitations of conventional diagnostic and treatment approaches in thyroid disorders highlight the necessity for innovative solutions. Artificial intelligence (AI) and machine learning (ML) technologies offer promising avenues to address these challenges. AI-driven tools can automate image analysis, reducing inter-observer variability in ultrasound interpretations. Predictive analytics can assist in risk stratification, identifying patients at higher risk of malignancy or recurrence. Furthermore, AI algorithms can optimize levothyroxine dosing by analyzing patient-specific data, leading to more personalized and effective treatment plans. By integrating these technologies into clinical practice, healthcare providers can enhance diagnostic accuracy, streamline treatment planning, and improve patient outcomes in thyroid disorder management.

Advertisement

3. AI techniques in thyroid disorder diagnosis

Artificial intelligence (AI) has revolutionized the field of thyroid disorder diagnosis by enhancing accuracy, reducing subjectivity, and streamlining clinical workflows. This section explores the key AI techniques employed in thyroid diagnostics, including machine-learning approaches, multimodal diagnostic integration, and real-world clinical implementation case studies.

3.1 Machine learning approaches

Machine learning (ML), a subset of AI, has been instrumental in automating the classification and diagnosis of thyroid disorders. Algorithms such as decision trees, support vector machines (SVMs), random forests, and deep learning models like convolutional neural networks (CNNs) have accurately differentiated benign and malignant thyroid nodules. For instance, a study reported that random forest classifiers achieved an accuracy of 99.81% in predicting hypothyroidism, outperforming other algorithms like k-nearest neighbors (KNN) and SVM [8].

Deep learning models, particularly CNNs, have shown remarkable performance in analyzing ultrasound images of thyroid nodules as shown in Figure 1. These models can automatically extract relevant features from imaging data, reducing the reliance on manual interpretation and minimizing inter-observer variability. In one study, a deep learning algorithm achieved an area under the curve (AUC) of 0.69, comparable to experienced radiologists, in classifying thyroid nodules on ultrasound images [9]. Deep learning models have been developed that have successfully identified lymph nodes involved in thyroid cancers.

Figure 1.

Application of artificial intelligence in thyroid surgery.

The data standards of the clinical characteristics used to train these models are of ample significance. They undergo data cleaning, for example, removal of duplicate values and missing variables. The datasets are then divided into testing, training, and validation sets. The training set is used to teach the model, the testing set checks how well the model learned, and the validation set helps adjust the model’s settings.

3.2 Multimodal diagnostic integration

Integrating data from multiple diagnostic modalities enhances the accuracy and reliability of thyroid disorder diagnosis. AI systems can combine information from ultrasound imaging, fine-needle aspiration (FNA) cytology, thyroid function tests (TFTs), and patient clinical data to provide a comprehensive assessment. This multimodal approach allows for more precise risk stratification and personalized treatment planning.

For example, AI-based computer-aided diagnosis (CAD) systems have been created to interpret ultrasound features, which has resulted in a decrease in unnecessary FNAs. Additionally, AI algorithms can analyze electronic medical records and other clinical data, including but not limited to US features, such as composition, shape, margin, echogenicity, and calcifications, to identify patterns and risk factors for thyroid disorders, facilitating earlier detection and intervention [8].

However, their specificities in ruling out malignancies are lower than those of experienced radiologists. Researchers analyzing the methodology of these models for diagnostic purposes have pointed out some points that may have led to the low specificity and accuracy of the results using machine learning and deep learning techniques.

  • The studies had limited comparability and reproducibility due to different imaging protocols, segmentation methods, and scanners/vendors.

  • Their studies had variations in feature type, selection, and classifiers.

  • Most studies lacked adequate test and validation datasets and were developed using relatively small sample sizes.

Nonetheless, this still underscores the usefulness of the current CADs as they could be better used as screening tools for the objective [10].

3.3 Clinical implementation case studies

Several clinical studies have demonstrated the successful implementation of AI techniques in thyroid disorder diagnosis. In one case, a deep learning-based AI model called ThyNet was developed to differentiate between malignant tumors and benign thyroid nodules, showing promising results in assisting radiologists [11].

Another study focused on enhancing thyroid volumetry used by nuclear medicine specialists to determine the dosage of radioactive iodine for monitoring and treating various thyroid diseases, particularly for radioiodine therapy in hyperthyroidism. This was achieved by combining tracked 3D ultrasound with a deep neural network-based model for automatic segmentation. This approach significantly reduced inter-observer variability and increased the accuracy of thyroid volume measurements, leading to more consistent assessments; however, this model was also trained using a low number of volunteers [12]. The improved performance can be attributed to the nature of the 3D acquisition and the automatic segmentation. The study’s small, healthy participant group and equipment specificity limit generalizability.

These case studies demonstrate AI’s potential to enhance diagnostic accuracy, reduce variability, and improve clinical outcomes in managing thyroid disorders. Larger, diverse samples and long-term studies are needed to validate and integrate the technique into clinical practice.

Advertisement

4. AI in treatment planning and management

4.1 Risk stratification and prognostication

Determining the best course of treatment for thyroid disorders requires accurate risk assessment. Artificial intelligence (AI) models, especially machine learning algorithms, have proven their worth in successfully classifying patients according to the probability of disease progression or recurrence. To help identify patients who might benefit from more aggressive interventions or closer surveillance, predictive models that use clinicopathological features, for example, have demonstrated high accuracy in predicting the recurrence of differentiated thyroid cancer.

4.2 Personalized therapeutic decision-making

AI facilitates the development of individualized treatment plans by integrating diverse data sources, including genetic profiles, imaging studies, and laboratory results. In thyroid cancer, AI algorithms assist in tailoring the most appropriate therapeutic modalities like surgery, radioactive iodine therapy, or active surveillance based on tumor characteristics and patient preferences. Moreover, achieving clinical and biochemical euthyroidism is difficult, and movement beyond traditional weight-based dosing strategies becomes necessary to achieve more personalized and precise medical treatment.

AI-driven tools can optimize levothyroxine dosing in hypothyroid patients by analyzing diverse clinical factors such as age, weight, comorbidities, and response to previous treatments, enhancing therapeutic efficacy and minimizing adverse effects. Machine learning models (Extra Trees Regressor) demonstrated the highest accuracy (R2 of 87.37%) and the lowest mean absolute error (MAE of 9.4 mcg) in predicting dosing requirements. BMI emerged as the strongest predictor of levothyroxine replacement therapy. These findings of a retrospective study indicate that using models in clinical systems could help personalize dosing for hypothyroidism; however, the study’s design and missing adherence data are major drawbacks, so future studies should focus on improving dosing strategies and treatment outcomes [13].

4.3 Monitoring and follow-up

Digital therapeutics (DTx) and Remote Continuous Data Monitoring (RCDM) are essential for managing chronic thyroid conditions and detecting potential complications. AI-powered applications and AI-assisted wearable devices enable real-time tracking of physiological parameters, such as heart rate, sleep patterns, and weight fluctuations, providing valuable insights into a patient’s response to therapy. These technologies support proactive adjustments to treatment regimens and facilitate timely interventions, ultimately improving patient adherence and outcomes.

4.4 Enhancing treatment planning and outcomes

AI contributes to preoperative planning in surgical management by analyzing imaging data to delineate anatomical structures and identify potential challenges. Advanced AI models, such as the E2E-Swin-Unet++, have been developed to enhance the precision of procedures like radiofrequency ablation by providing real-time segmentation of thyroid tumors, thereby improving targeting accuracy and reducing the risk of complications. In a world of precision medicine, imaging investigations are evaluated to facilitate surgical planning more efficiently. Real-time AI-assisted intraoperative guidance can prospectively transform robotic-assisted thyroid surgery. AI models have been devised based on endoscopic thyroid surgery datasets to identify the Recurrent Laryngeal Nerve (RLN), giving rise to promising results as in Table 1.

S. noAI technologyFunction in thyroid surgeryKey applicationsPotential benefits
1.Machine learningPredicts post-operative complicationsRisk modeling for nerve injury, hypocalcemia, and vocal cord issuesImproved surgical planning, reduced complications
2.Deep learningClassifies thyroid tumors as benign or malignantHistopathology image analysis, cytology interpretationFaster and more accurate diagnosis, aiding treatment decisions
3.Natural language processing (NLP)Analyzes patient feedback and surgical notesExtracts insights from EHRs, follow-up notes, and patient reportsIdentifies care gaps, and improves documentation and outcomes
4.Augmented reality (AR)Provides 3D surgical navigation through AR overlaysVisualizes critical structures (e.g., nerves, vessels) during surgeryEnhances precision, reduces errors during operations
5.Computer visionIdentifies anatomical structures in real-time during surgeryReal-time recognition of glands, nerves, blood vesselsSupports intraoperative decision-making

Table 1.

AI in thyroid disorders.

Future research should concentrate on continuous learning mechanisms for AI models, utilizing periodically updated data and feedback from clinical use to improve the accuracy and relevance of these models over time. Exploring hybrid models that combine convolutional neural networks with transformers and integrating multi-modal learning approaches that utilize different forms of clinical data may further enhance the model’s accuracy and robustness [14].

Ultrasound-guided thoracoscopic surgery based on an artificial intelligence algorithm has significantly better outcomes in terms of postoperative pain and reduced surgical time, as it offers better image resolution compared to conventional methods. This approach also maintains high safety standards and provides substantial clinical benefits, although further research is needed to confirm long-term efficacy [15].

4.5 AI innovations in robotic thyroid surgery

AI enhances robotic thyroid surgery by optimizing preoperative planning, real-time decision support, and post-operative tracking, improving precision, safety, and patient outcomes. This revolutionizes the preoperative planning process by allowing precise mapping of the vital structures related to the thyroid gland. Additionally, AI assists in robotic-assisted thyroid surgery, optimizing instrument movements, predicting potential complications, and minimizing tissue damage. Machine learning algorithms analyze surgical videos to identify best practices and offer an adequate assessment of resident training [16]. Furthermore, AI-driven simulation platforms provide immersive training environments, allowing surgeons to practice procedures in virtual reality before performing them on patients aiding in skill development and procedural efficiency.

4.6 Integration into multidisciplinary care

AI systems can streamline multidisciplinary team (MDT) meetings by aggregating and presenting relevant patient data, facilitating collaborative decision-making. By offering evidence-based recommendations and highlighting critical information, AI fosters efficient and effective discussions among endocrinologists, surgeons, radiologists, and other healthcare professionals involved in thyroid disorder management. Studies have proven enhanced concordance rates of LLM-produced results with guidelines and MDT recommendations; however, large prospective multicenter data must be analyzed, and nevertheless, the importance of counterchecking the AI-generated response by experienced clinicians cannot be undermined. Transcription software can be utilized during MDT meetings to avoid missing important recommendations.

Advertisement

5. AI in thyroid cancer detection and treatment

Artificial intelligence (AI) has emerged as a transformative force in the detection and treatment of thyroid cancer, offering enhanced diagnostic accuracy, personalized treatment planning, and improved postoperative care. This section delves into the multifaceted applications of AI in thyroid oncology, highlighting its role in early detection, surgical planning, and postoperative management.

5.1 Early detection

Early and accurate detection of thyroid cancer is crucial for effective treatment and improved patient outcomes. AI technologies, particularly machine learning (ML) and deep learning (DL) algorithms, have shown remarkable proficiency in analyzing complex datasets to identify malignancies at an early stage.

5.1.1 Ultrasound imaging analysis

Ultrasound is a primary modality for evaluating thyroid nodules. AI-enhanced ultrasound analysis utilizes DL algorithms, such as convolutional neural networks (CNNs), to assess sonographic features like echogenicity, margins, and calcifications. These models have demonstrated diagnostic accuracies comparable to experienced radiologists, thereby aiding in the differentiation between benign and malignant nodules.

5.1.2 Cytopathological assessment

Fine-needle aspiration biopsy (FNAB) is a standard procedure for evaluating thyroid nodules. AI applications in cytopathology involve the use of DL models to analyze digitized cytological slides, enhancing the detection of malignant cells and reducing inter-observer variability. This approach streamlines the diagnostic process and assists pathologists in making more accurate assessments.

5.1.3 Molecular marker integration

AI models can integrate molecular and genetic data, such as BRAF and RAS mutations, to assess the risk of malignancy in thyroid nodules. By combining imaging findings with molecular profiles, AI facilitates a more comprehensive risk stratification, guiding clinicians in decision-making regarding the necessity of surgical intervention.

5.2 Surgical planning and postoperative care

AI’s role extends beyond diagnosis to encompass surgical planning and postoperative management, ensuring precision and personalized care.

5.2.1 Preoperative planning

Accurate delineation of tumor boundaries and identification of critical anatomical structures are paramount for successful thyroid surgery. AI-driven image segmentation tools, employing architectures like U-Net, assist surgeons in visualizing the tumor extent and planning the surgical approach, thereby minimizing complications and preserving vital structures. U-Net is a CNN structure widely used in deep learning, mainly for image segmentation [17].

5.2.2 Postoperative monitoring

Monitoring for recurrence is a critical aspect of postoperative care. AI algorithms can analyze serial ultrasound images to detect subtle changes indicative of recurrent disease. Additionally, AI models can predict recurrence risk by evaluating postoperative histopathological data and patient-specific factors, enabling timely interventions.

5.3 Case studies

The Mayo Clinic has pioneered the development of a novel class of AI, termed hypothesis-driven AI, which integrates existing medical knowledge with data-driven approaches. This AI model has demonstrated a 15% improvement in thyroid cancer detection rates, underscoring its potential to enhance diagnostic accuracy and inform treatment strategies.

In another instance, researchers at the Mayo Clinic developed high-resolution ultrasound imaging software compatible with existing ultrasound machines. This technology, termed quantitative high-definition microvessel imaging (q-HDMI), albeit investigative, captures high-resolution 2D and 3D images combined with AI algorithms, thus allowing the visualization of microvessels to the size of human hair within tumors, studying their distinct morphological feature called biomarkers that are out of the scope of this chapter, facilitating earlier and more accurate detection of malignancies [18].

These case studies exemplify the transformative impact of AI in thyroid cancer care, highlighting its capacity to augment clinical decision-making and improve patient outcomes.

Advertisement

6. Ethical and legal considerations

The integration of artificial intelligence (AI) into the diagnosis and management of thyroid disorders presents significant ethical and legal challenges. These challenges range across algorithmic bias, data privacy, and evolving regulatory frameworks. Addressing these issues is crucial to ensure equitable, safe, and compliant AI applications in thyroid care.

6.1 Bias and fairness

AI systems in healthcare are susceptible to biases that carry the potential to worsen existing health disparities. Non-diverse training datasets lead to misdiagnoses or suboptimal treatment recommendations for underrepresented groups. For instance, studies have shown that AI models may misdiagnose patients of color due to underrepresentation in training data. Hallucination or confabulation is another term given to the phenomenon where an AI model might hallucinate to make a response of its own if it is not trained on the relevant data.

Some of the measures to mitigate these biases are:

  • Ensure diverse training data: Incorporate data from varied demographic groups to enhance model generalizability.

  • Implement bias audits: Regularly assess AI systems for discriminatory outcomes and rectify identified issues.

  • Promote transparency: Maintain openness in AI development processes to facilitate accountability and trust.

By proactively addressing bias, AI can be harnessed to improve thyroid disorder management across diverse populations [19].

6.2 Data privacy and security

The deployment of AI in healthcare warrants rigorous data privacy and security procedures. Compliance with regulations such as the Health Insurance Portability and Accountability Act (HIPAA) in the U.S. and the General Data Protection Regulation (GDPR) in the EU is crucial.

Key considerations include:

  • Data encryption: Protect patient data through encryption during storage and transmission.

  • Access controls: Implement strict access protocols to ensure that only authorized personnel can access sensitive information.

  • Consent management: Explicit patient consent should be signed for data usage, aligning with GDPR’s emphasis on informed consent.

Adhering to these practices not only ensures legal compliance but also fosters patient trust in AI-driven healthcare solutions.

6.3 Regulatory frameworks

Regulatory bodies are evolving to address the unique challenges posed by AI in medical devices. In the United States, the Food and Drug Administration (FDA) has issued draft guidance emphasizing a total product lifecycle approach for AI-enabled medical devices.

6.3.1 Key aspects of the FDA’s approach include

  • Premarket evaluation: Assessing AI devices through pathways like premarket clearance (510(k)), De Novo classification, or premarket approval, depending on risk levels put forward by the United States Food and Drug Administration

  • Postmarket monitoring: Continuous oversight to ensure ongoing safety and effectiveness of AI systems.

  • Transparency requirements: Mandating clear documentation of AI algorithms’ functionality and decision-making processes.

These regulatory measures aim to balance innovation with patient safety, ensuring that AI applications in thyroid disorder management are both effective and trustworthy.

Advertisement

7. Conclusion

We are standing at a fascinating crossroads in thyroid medicine. The arrival of AI is not just changing our tools—it is reshaping how we think about patient care altogether. Imagine a world where diagnoses are sharper, treatments fit like custom-tailored suits, and healthcare resources flow exactly where they are needed most. That’s the promise AI is already starting to deliver in thyroid clinics worldwide.

Bringing these futuristic tools into real exam rooms is not as simple as flipping a switch. Doctors and patients alike face real questions: How do these algorithms reach their conclusions? Can we trust them with sensitive health data? Will they work as well for a grandmother in rural Africa as they do for a young executive in New York or Tokyo? Training in AI is a requirement of the modern world for endocrinologists, surgeons, and radiologists alike.

The road ahead has some clear signposts. We need AI systems that can explain their thinking in plain language, not just spit out recommendations. The development of reasoning models in recent times is a step toward that. They must prove their worth across all communities, not just in high-tech research hospitals. And perhaps most importantly, they need to slide effortlessly into the existing rhythms of healthcare, not force doctors to learn entirely new ways of working.

It is exciting to watch AI grow from a helpful assistant to a true partner in care. It is not about replacing the doctor’s intuition, but about combining human wisdom with machine precision. Here comes the “human-in-the-loop feedback,” which implies human involvement in the form of reinforcement learning to give results that are enhanced and more toward artificial general intelligence than AI alone. The thyroid patients of tomorrow may never know about the complex algorithms working behind the scenes—they will just experience care that’s more accurate, more personal, and more responsive than we ever thought possible [20].

This transformation would not happen overnight, but each month brings breakthroughs. As someone who has witnessed both the frustrations of traditional thyroid care and the potential of these new technologies, I cannot help but feel optimistic. The future of thyroid medicine is not cold and computerized—it is warmer, smarter, and more human than ever before.

Advertisement

Acknowledgments

The authors would like to express sincere gratitude to the following:

  1. Rodolfo J. Oviedo, MD, FACS, FASMBS, Medical Director of Bariatric Surgical Services and Robotics Program at Nacogdoches Medical Center in Nacogdoches, Texas, USA, for his guidance and encouragement throughout the preparation of this chapter, including the collaboration between the authors.

  2. Irfan Rizvi, MD, FACS, FASCRS, FRCS, Chairman, Department of Surgery, Virginia Hospital Center, Assistant Professor of Surgery, Uniformed Services University of Health Sciences, for his insightful suggestions that helped define the tone and scope of the manuscript.

  3. Dr. Sheraz Naseer, PhD, Artificial Intelligence, Data Science, Chairperson, Department of Artificial Intelligence, University of Management and Technology—UMT Lahore, Pakistan, for his contributions in clarifying and deepening our understanding of the core concepts and applications of artificial intelligence and its various subfields.

Their collective support and input greatly enhanced the quality and direction of this work.

References

  1. 1. Atighi F, Yazdanpanahi P, Keshtkar A, Karimi A, Naseri A, Dabbaghmanesh MH. Illuminating the path to thyroid disorder management using artificial intelligence: A narrative review. Shiraz E-Medical Journal. 2024;26(2):1-12. DOI: 10.5812/semj-e151031
  2. 2. Kaul V, Enslin S, Gross SA. History of artificial intelligence in medicine. Gastrointestinal Endoscopy. 2020;92(4):807-812. DOI: 10.1016/j.gie.2020.06.040. Epub 2020 Jun 18
  3. 3. Toro-Tobon D, Loor-Torres R, Duran M, et al. Artificial intelligence in thyroidology: A narrative review of the current applications, associated challenges, and future directions. Thyroid. 2023;33(8):903-917. DOI: 10.1089/thy.2023.0132
  4. 4. Alyami J, Almutairi FF, Aldoassary S, et al. Interobserver variability in ultrasound assessment of thyroid nodules. Medicine (Baltimore). 2022;101(41):e31106. DOI: 10.1097/MD.0000000000031106
  5. 5. de Jong MC, McNamara J, Winter L, Roskell D, Khan S, Mihai R. Risk of malignancy in thyroid nodules with indeterminate (THY3f) cytology. Annals of the Royal College of Surgeons of England. 2022;104(9):703-709. DOI: 10.1308/rcsann.2021.0358
  6. 6. Meyer AND, Murphy DR, Al-Mutairi A, et al. Electronic detection of delayed test result follow-up in patients with hypothyroidism. Journal of General Internal Medicine. 2017;32(7):753-759. DOI: 10.1007/s11606-017-3988-z
  7. 7. Gupta P, Rustam F, Kanwal K, et al. Detecting thyroid disease using an optimized machine learning model based on differential evolution. International Journal of Computational Intelligence Systems. 2024;17(3):1-19. DOI: 10.1007/ s44196-023-00388-2
  8. 8. Weng J, Wildman-Tobriner B, Buda M, et al. Deep learning for classification of thyroid nodules on ultrasound: Validation on an independent dataset. Clinical Imaging. 2023;99:60-66. DOI: 10.1016/j.clinimag.2023.04.010
  9. 9. Ahmad MA-S, Haddad J. An explainable AI model for predicting the recurrence of differentiated thyroid cancer. In: Second Jordanian International Biomedical Engineering Conference (JIBEC); Amman, Jordan. Vol. 2024. New Jersey, USA: IEEE; 2024. pp. 84-89. DOI: 10.1109/JIBEC63210.2024.10932125
  10. 10. Ha EJ, Baek JH. Applications of machine learning and deep learning to thyroid imaging: Where do we stand? Ultrasonography. 2021;40(1):23-29. DOI: 10.14366/usg.20068
  11. 11. Nagendra L, Pappachan JM, Fernandez CJ. Artificial intelligence in the diagnosis of thyroid cancer: Recent advances and future directions. Artificial Intelligence in Cancer. 2023;4(1):1-10. DOI: 10.35713/aic.v4.i1.1
  12. 12. Krönke M, Eilers C, Dimova D, et al. Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry. PLoS ONE. 2022;17(7):e0268550. DOI: 10.1371/journal.pone.0268550
  13. 13. Ngan TT, Tra DH, Mai NTQ, Dung HV, Khai NV, Linh PV, et al. Developing a machine learning-based predictive model for levothyroxine dosage estimation in hypothyroid patients: A retrospective study. Frontiers in Endocrinology. 2025;16:1-14. DOI: 10.3389/fendo.2025.1415206
  14. 14. Dialameh M, Rajabzadeh H, Sadeghi-Goughari M, Sim JS, Kwon HJ. E2E-Swin-Unet++: An enhanced end-to-end Swin-Unet architecture with dual decoders For PTMC segmentation [preprint]. arXiv. 2024. Available from: https://arxiv.org/abs/2410.18239
  15. 15. Shen X, Yuan A, Zhang K. Ultrasound image under artificial intelligence algorithm in thoracoscopic surgery for papillary thyroid carcinoma. Scientific Programming. 2022;2022(1):2646094
  16. 16. Park J, Kim K. Current and future of robotic surgery in thyroid cancer treatment. Cancers. 2024;16(13):2470. DOI: 10.3390/cancers16132470
  17. 17. Yang L, Wang X, Zhang S, Cao K, Yang J. Research progress on artificial intelligence technology-assisted diagnosis of thyroid diseases. Frontiers in Oncology. 2025;15:1536039. DOI: 10.3389/fonc.2025.1536039
  18. 18. Kurti M, Sabeti S, Robinson KA, Scalise L, Larson NB, Fatemi M, et al. Quantitative biomarkers derived from a novel contrast-free ultrasound high-definition microvessel imaging for distinguishing thyroid nodules. Cancers. 2023;15(6):1888. DOI: 10.3390/cancers15061888
  19. 19. Chang T, Nuppnau M, He Y, Kocher KE, Valley TS, Sjoding MW, et al. Racial differences in laboratory testing as a potential mechanism for bias in AI: A matched cohort analysis in emergency department visits. PLOS Global Public Health. 2024;4(10):e0003555. DOI: 10.1371/journal.pgph.0003555
  20. 20. Mosqueira-Rey E, Hernández-Pereira E, Alonso-Ríos D, et al. Human-in-the-loop machine learning: A state of the art. Artificial Intelligence Review. 2023;56:3005-3054. DOI: 10.1007/s10462-022-10246-w

Written By

Qaviullah Mian and Nebiyou Bayleyegn

Submitted: 16 April 2025 Reviewed: 05 May 2025 Published: 09 July 2025