Computer-aided diagnosis of pectus excavatum using CT images and deep learning methods

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computer-Aided Diagnosis

Background:

• Pectus excavatum (PE) is a common chest wall defect requiring accurate assessment for surgical planning.
• Current index-based evaluations of PE lack comprehensive information from CT images and can be prone to errors due to individual variations.
• Objective and reliable methods are needed to overcome the limitations of existing PE assessment techniques.

Purpose of the Study:

• To develop a computer-aided diagnosis (CAD) system utilizing convolutional neural networks (CNNs) for automated classification of PE from chest CT images.
• To enhance diagnostic accuracy and overcome limitations of traditional index-based PE assessments.
• To explore transfer learning and block-wise fine-tuning strategies to mitigate overfitting with limited datasets.

Main Methods:

• Development of a CNN-based CAD system for automatic feature learning and PE image classification.
• Application of block-wise fine-tuning with transfer learning to optimize model performance and reduce overfitting.
• Implementation of a majority rule-based voting method for integrated, comprehensive patient-level diagnostic results.

Main Results:

• The proposed CNN-based CAD system achieved a high classification accuracy of 94.76% for PE diagnosis.
• Block-wise fine-tuning strategies were explored to determine optimal transfer learning parameters.
• A majority rule-based voting method successfully integrated classification results for comprehensive thoracic assessment.

Conclusions:

• The CNN-based CAD system demonstrates significant potential for accurate and automated diagnosis of pectus excavatum.
• The developed system offers a more comprehensive assessment compared to traditional index-based methods.
• This approach paves the way for improved clinical evaluation and management of PE.