Patient-level thyroid cancer classification using attention multiple instance learning on fused multi-scale...

Area of Science:

• Medical imaging
• Artificial intelligence in healthcare
• Oncology

Background:

• Accurate diagnosis of malignant thyroid nodules is crucial for treatment planning.
• Ultrasound image assessment and current deep learning models face challenges with nodule segmentation and image magnification variability.
• Biopsy is often required for definitive diagnosis, despite limitations in ultrasound assessments.

Purpose of the Study:

• To develop an advanced deep learning model for accurate patient-level malignancy classification of thyroid nodules using ultrasound images.
• To overcome limitations of existing methods dependent on manual segmentation and handle image magnification heterogeneity.

Main Methods:

• Developed a multi-scale, attention-based multiple-instance learning model.
• Fused global and local features from diverse ultrasound frames.
• Utilized patient-level malignancy classification approach.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.785 and an Area Under the Precision-Recall Curve (AUPRC) of 0.539.
• Significantly outperformed a baseline model (AUROC: 0.667, AUPRC: 0.444) trained on clinical features.
• Demonstrated improved classification performance for triaging biopsy necessity.

Conclusions:

• The novel deep learning model enhances the accuracy of thyroid nodule malignancy classification from ultrasound images.
• This approach offers a more reliable method for diagnosing thyroid nodules, potentially reducing invasive procedures.
• Improved diagnostic capabilities can lead to more efficient patient management and treatment strategies.