Rapid and accurate clinical diagnosis using mass spectrometry (MS) data is challenging due to the need for time-consuming preprocessing steps in conventional machine learning approaches. This study investigates the use of convolutional neural networks (CNNs) for MS data classification, focusing on transfer learning and a novel cumulative learning method to address the issue of limited training samples in medical applications. By training a single model on multiple small datasets from various biological contexts, the cumulative learning approach achieves over 98% classification accuracy for 1D clinical MS data, demonstrating its potential for improving the accuracy and speed of MS-based clinical diagnosis.

Khawla Seddiki, Philippe Saudemont, Frédéric Precioso, Nina Ogrinc, Maxence Wisztorski, Michel Salzet, Isabelle Fournier, Arnaud Droit