Smartphone-enabled AI system shows high accuracy in detecting papillary thyroid cancer

A new study published in Thyroid Science reports that a smartphone-enabled artificial intelligence (AI) system can accurately detect papillary thyroid carcinoma (PTC) from ultrasound images, offering a potentially low-cost and secure diagnostic tool for broader clinical use.

PTC accounts for approximately 90% of all thyroid cancers, and ultrasound remains a cornerstone of diagnosis. However, diagnostic accuracy often depends on specialist expertise, which is not uniformly available across healthcare settings. To address this gap, investigators developed an AI-based system that integrates a standalone server with a smartphone application capable of analyzing images captured directly from an ultrasound monitor.

The system was trained using ultrasound images from 150 PTC cases and 200 benign thyroid lesions. In validation testing, the AI server achieved a sensitivity of 100%, specificity of 96.7%, and overall accuracy of 97.8%. When used via the smartphone interface, performance remained high, with a sensitivity of 100%, specificity of 90.0%, and overall accuracy of 93.3%.

Only one misclassification occurred during testing, in which tracheal cartilage was incorrectly identified as malignant.

The system uses an object detection model (YOLOv5) that allows it to identify and classify lesions directly from raw ultrasound images without requiring manual cropping. This design supports real-world clinical workflows and distinguishes it from some existing AI tools that require additional preprocessing.

The authors emphasized that the system was designed to overcome two major barriers to AI adoption in clinical practice: cost and data security. Unlike many AI platforms that require integration with hospital networks or proprietary imaging systems, this approach uses a standalone server and does not store patient data. All data transmissions are encrypted, and the system operates independently of hospital information systems.

“The developed AI system enables highly accurate diagnosis of PTC from ultrasound images,” the study noted. “The smartphone interface retains robust diagnostic performance, offering a low-cost, secure, and accessible tool for clinical use, particularly in resource-limited settings.”

Democratizing High-Quality Medical Care

Lead author Yoshihiro Takahashi, MD, also provided additional commentary in an interview with AACE Endocrine AI to provide clinicians practical information on AI in Endocrinology, and highlighted the broader implications of the findings, particularly for healthcare access.

“I believe the true value of this smartphone-enabled AI system lies in its potential to democratize high-quality medical care,” he said. “Our system empowers non-specialist physicians in rural or small clinics to achieve specialist-level diagnostic accuracy simply by taking a picture of the ultrasound monitor with a smartphone."

Dr Takahashi also underscored the importance of affordability and cybersecurity, noting that they achieved a “low-cost, high-security, and highly portable” solution that doesn’t expose hospital information systems to cyber threats. Additionally, the system is designed for seamless integration into existing clinical workflows.

“Because our object-detection-based AI can analyze raw images captured directly from the screen without the need for manual image cropping or expensive hardware upgrades, it seamlessly integrates into the daily workflow of frontline doctors,” he noted.

Despite promising results, the study has limitations. The training dataset was relatively small, and the model was developed exclusively for PTC, limiting its ability to detect other types of thyroid malignancies. The authors indicated that future work will expand the dataset and broaden the system’s diagnostic capabilities.

Overall, however, the findings suggest that smartphone-enabled AI could play a significant role in improving access to accurate thyroid cancer diagnosis, particularly in underserved or resource-limited settings.

The authors reported no known competing interests.

AACE Endocrine AI is published by Conexiant under a license arrangement with the American Association of Clinical Endocrinology, Inc. (AACE^®). The ideas and opinions expressed in AACE Endocrine AI do not necessarily reflect those of Conexiant or AACE. For more information, see Policies.