Clinical Scorecard: Multi-Class Classification of Thyroid Disorders Utilizing Modified DenseNet-201 and Tc-99m Scintigraphy Through Deep Learning Techniques

At a Glance

Key Highlights

• First detailed classification of seven thyroid disorders using modified DenseNet-201 CNN on Tc-99m scintigraphy images.
• Use of a real-world dataset of 1599 thyroid scintigraphy scans collected over 10 years from BINOR Cancer Hospital.
• Modified DenseNet-201 model improves feature extraction and classification from low-resolution scintigraphy images aligned with nuclear medicine workflows.

Guideline-Based Recommendations

Diagnosis

• Combine clinical assessment, high-resolution ultrasound, serum TSH quantification, and fine-needle aspiration biopsy for definitive diagnosis.
• Utilize Tc-99m thyroid scintigraphy for physiological and functional assessment of the thyroid gland.
• Apply AI-based deep learning models such as modified DenseNet-201 CNN to improve classification accuracy of thyroid disorders from scintigraphy images.

Management

• Tailor treatment strategies based on accurate classification of thyroid disorder severity and type.
• Avoid overtreatment or undertreatment by improving diagnostic precision through AI-assisted imaging analysis.

Monitoring & Follow-up

• Monitor thyroid function and nodule characteristics using combined imaging and biochemical tests.
• Use AI tools to reduce inter- and intra-observer variability in image interpretation.

Risks

• Be aware of limitations of scintigraphy including lower image resolution and potential human error in manual analysis.
• Consider the risk of diagnostic ambiguity leading to inappropriate clinical decisions without AI assistance.
• Recognize limited availability of nuclear medicine physicians and potential fatigue impacting diagnosis quality.

Patient & Prescribing Data

Patients with suspected thyroid nodules or thyroid gland dysfunction undergoing scintigraphy imaging.

Accurate multi-class classification of thyroid disorders via AI models can guide appropriate treatment decisions and improve patient outcomes.

Clinical Best Practices

• Integrate AI-based deep learning models with conventional diagnostic methods to enhance accuracy.
• Use Tc-99m scintigraphy as a functional imaging modality complementary to anatomical imaging.
• Ensure multidisciplinary evaluation including clinical, biochemical, cytological, and imaging data for comprehensive diagnosis.
• Address inter- and intra-observer variability by adopting automated AI interpretation tools.
• Maintain awareness of AI model limitations and validate findings with clinical correlation.

References

• Thyroid gland function and disorders overview
• Role of Tc-99m scintigraphy in thyroid assessment
• Deep learning in thyroid nodule classification
• Artificial intelligence applications in thyroid disease diagnosis