Machine learning for predicting young's modulus of soft tissues from atomic force microscopy data

Atomic force microscopy (AFM) is an advanced tool applied to investigate the mechanical properties of a wide variety of material classes, including soft biological tissues. Force-indentation curves obtained through AFM can be analyzed using appropriate contact models to extract key material parameters, notably Young's modulus. The measurement of the Young's modulus of biological materials imposes significant constraints. However, a persistent challenge in the adoption of AFM lies in the complexity of data analysis, which is often time-intensive and requires highly specialized personnel with robust expertise in physics and mathematics. As a result, recent efforts have concentrated on alternative methodologies, particularly machine learning (ML) models, to enhance and facilitate this analysis process. This work introduces a customized regressor to predict Young's modulus from force-indentation (F-I) curves, specifically targeting values between 1 and 50 kPa. The model was exclusively trained on synthetic datasets generated based the Johnson-Kendall-Roberts (JKR) contact theory. An MLP regressor was employed for model training, alongside a Grid-Search library to optimize the hyperparameters. Model accuracy was validated through regressor testing against theoretical and laboratory force-indentation curves obtained from AFM nano-indentation studies. By delivering precise predictions with optimized computational efficiency, this regressor demonstrates the potential of deep learning approaches to support comprehensive evaluation of AFM nanoindentations, paving the way for further advancements in overcoming the current limitations associated with AFM.