Deep Learning-Assisted Detection and Classification of Thymoma Tumors in CT Scans

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thymoma, a rare thymus gland neoplasm, presents diagnostic challenges on CT scans due to subtle morphology.
• Accurate differentiation from other mediastinal pathologies and subtype classification (benign vs. malignant) is crucial.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for enhanced thymoma detection and classification.
• To improve diagnostic accuracy for distinguishing thymoma from healthy cases and classifying thymoma subtypes.

Main Methods:

• A hybrid DL model integrating VGG16 for feature extraction and MLP-Mixer for feature enhancement was developed.
• Customized image preprocessing and post-processing techniques were applied.
• Performance was benchmarked against state-of-the-art DL models using accuracy, F1 score, recall, and precision.

Main Results:

• The proposed model achieved 97.15% accuracy and 80.99% F1 score for thymoma vs. healthy classification.
• For benign vs. malignant thymoma classification, the model reached 79.20% accuracy and 78.51% F1 score.
• The model outperformed all baseline methods in both classification tasks.

Conclusions:

• The combined VGG16 and MLP-Mixer approach offers superior and balanced performance for thymoma diagnosis.
• This DL model shows significant potential for clinical decision support in identifying and classifying thymoma.