Head and neck cancer ranks as the seventh most common cancer globally. In Taiwan, approximately 8,000 new cases are reported annually. Traditionally, diagnosing and treating head and neck cancer rely heavily on the expertise of radiologists and radiation oncologists, requiring time-intensive manual interpretation and annotation. This process is not only laborious but also subject to variability due to human factors, potentially leading to inconsistent diagnoses.

To address these challenges, a team led by Shinn-Ying Ho, Distinguished Professor at National Yang Ming Chiao Tung University, collaborated with Dr. Ying-Yen Huang and Dr. Tzu-Ting Huang from Kaohsiung Chang Gung Memorial Hospital to develop the world’s first AI-powered clinical diagnostic assistance system for head and neck cancer. This system integrates two core functions: automatic lymph node lesion annotation and prediction of extracapsular spread (ECS) of lymph nodes. It significantly improves the efficiency of diagnosis and annotation, marking a milestone in precision medicine.

Evolutionary Learning: Decoding the Medical Black Box

While deep learning models excel in automatically detecting and annotating lymph node regions in CT images of patients with head and neck squamous cell carcinoma, they often fall short in providing explainable results for lymph node metastasis and ECS. To enhance accuracy and transparency, Professor Ho’s team incorporated evolutionary learning technology into their AI system.

Advantages of Evolutionary Learning:

• Enhanced Explainability: Evolutionary learning optimizes the selection of key features to build personalized predictive models, providing insights into individual patient characteristics. This increases clinicians’ confidence in AI-assisted decision-making.
• Tackling Biomedical Data Challenges: Evolutionary learning is adept at handling unique challenges in biomedical imaging, such as high-dimensional data, limited sample sizes, strong feature interactions, and annotation uncertainties. It identifies robust features and creates optimized predictive models.
• Flexible Applications: Evolutionary learning is versatile, applicable to various tasks involving integrated analysis of images and data, including medical imaging, EEG/ECG signals, microbiomes, clinical data, and gene expression profiles, demonstrating broad clinical potential.

Revolutionary AI-Assisted Diagnosis System

The automatic annotation system leverages deep learning to quickly and accurately identify lesions in head and neck cancer images. It also allows physicians to refine annotations, ensuring precision and professionalism. The diagnostic assistance system employs innovative evolutionary learning, integrating 2D radiomics, 3D imaging, and clinical pathological data to construct predictive models. These models achieve over 80% accuracy and sensitivity in predicting ECS and metastasis, addressing the low sensitivity and diagnostic variability often faced by radiology experts. The system also provides interpretable features to support clinical decision-making, boosting confidence in its application.

Having undergone feasibility testing at Kaohsiung Chang Gung Memorial Hospital, the system continues to be optimized and clinically validated, advancing toward the goal of precision radiotherapy.

National Yang Ming Chiao Tung University’s AI Breakthrough

Professor Shinn-Ying Ho highlighted that traditional manual interpretation of a single case can take hours, while the team’s AI-assisted diagnostic system completes complex image analysis within seconds. This breakthrough not only enhances diagnostic efficiency but also provides more precise, personalized treatment information for patients with head and neck cancer, representing a significant leap forward in precision medicine.

He further emphasized that the success of this AI system stems from the team’s long-term commitment to artificial intelligence and medical imaging research. The multidisciplinary team, comprising clinical physicians, medical imaging experts, information scientists, and biostatisticians, overcame numerous technical challenges. Moving forward, the team aims to expand the system’s applications to other cancers, such as liver and gastric cancer, while developing a user-friendly interface to make the system more accessible to clinicians and promote widespread adoption of AI-assisted diagnostics.

Resource: AI演化學習輔助判讀 3分鐘診斷頭頸癌