Purpose: Convolutional Neural Network (CNN) is a powerful Deep Learning (DL) model used in healthcare for diseases classification through image analysis. Although primarily designed for image data, CNNs can be used with tabular data by converting it into images format. This work aims to assess the performance of CNNs in classifying cases of Congenital Syphilis (CS) using images generated from tabular data, and to compare their performance with Machine Learning (ML) models.
Methods: We compare different converters for transforming tabular data into images, selecting the most effective one based on the highest CNN metric outcomes. We perform hyperparameter optimization on the best CNN model using Random Search, and compare its performance against an optimized ML model, the Adaboost, proposed in a previous work.
Results: We find that it is possible to classify CS with images generated from tabular data of medical records. Additionally, after optimized, the CNN is capable of achieving competitive classification outcomes against an optimized Adaboost, implying their potential utility in medical data analysis.
Conclusion: This study demonstrates CNNs’ potential to classify CS cases using images from tabular medical data. The comparison with an optimized Adaboost showcases CNNs’ viability for medical data analysis, showcasing their adaptability beyond image analysis in structured healthcare data. The work limitations include the limited number of converters, the absence of a computational cost analysis, the focus on 2D CNNs only, and the absence of a rigorous statistical significance assessment.