Detection of Abnormal Changes on the Dorsal Tongue Surface Using Deep Learning

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Oral pathology diagnostics

Background:

• Tongue mucosa alterations are indicative of various local and systemic health conditions.
• Early detection of dorsal tongue abnormalities is crucial for timely diagnosis and treatment.
• Current diagnostic methods may benefit from advanced computational approaches.

Purpose of the Study:

• To evaluate the efficacy of deep learning (DL) for detecting abnormal regions on the dorsal tongue surface.
• To assess the performance of the VGG16 model in classifying normal and abnormal tongue images.
• To investigate the utility of point mapping segmentation for delineating abnormal tongue areas.

Main Methods:

• A dataset of 175 dorsal tongue images was collected and processed into 7782 cropped images.
• The VGG16 deep learning model was trained to classify image regions as normal, abnormal, or non-tongue.
• Point mapping segmentation was applied to 80 full-view dorsal tongue images for abnormal region analysis.

Main Results:

• The VGG16 model achieved high F1-scores for abnormal (0.960) and normal (0.968) classes in image prediction.
• Point mapping segmentation yielded average F1-scores of 0.727 for abnormal and 0.645 for normal regions.
• Segmentation evaluation demonstrated average precision of 0.940 for abnormal and 0.890 for normal areas.

Conclusions:

• The developed deep learning algorithm demonstrates high accuracy in identifying abnormal dorsal tongue surface areas.
• This AI-driven approach shows potential as a valuable tool to assist in the diagnosis of tongue mucosal diseases.
• The findings support the integration of deep learning into clinical practice for enhanced oral health diagnostics.