ChromoCheck: Predicting Postnatal Chromosomal Trisomy Cases Using a Support Vector Machine Learning Model

Chromosomal study via karyotype is one of the historical gold-standard procedures used to provide a clearer view of chromosomal trisomy abnormalities. It has been used to correlate several phenotypic manifestations that require immediate medical intervention. However, the laboratory procedure persisted with various drawbacks. The recent machine learning model shed light on prediction capabilities in the medical field. In this study, we aimed to use a support vector machine model for predicting postnatal chromosomal trisomy cases.

A dataset of 946 neonatal records from the Royal Hospital, Muscat, Oman, covering the period from 2013 to 2023, has been used in this model. The model is based on features such as thyroxine hormone levels and thyroid-stimulating hormone levels. With different R packages, we used a support vector machine model with leave-one-out cross-validation and ten iterations to test three kernel functions: linear, radial, and polynomial.

Among the obtained kernel performances, the linear kernel has optimal classification performance. The training accuracy was 81%, and the testing accuracy was 82%. Sensitivity ranged from 97 to 98%, and specificity ranged from 79 to 80%. The area under the curve in relation to the training dataset came to 0.89, and it came to 0.90 for the test dataset. We deployed the trained models in a website tool called ChromoCheck.

Our study is an example of how machine learning can be instrumental in augmenting conventional methods of cytogenetics diagnosis and decision-making in a clinical setup.