Comparative analysis of deep learning models for predicting biocompatibility in tissue scaffold images

Area of Science:

• Bioprinting and Tissue Engineering
• Artificial Intelligence in Medicine
• Biomaterials Science

Background:

• Bioprinting complex tissue scaffolds is crucial for tissue engineering.
• Predicting scaffold biocompatibility pre-fabrication is a significant challenge, leading to inefficiencies.
• Artificial Intelligence (AI) models, including ANNs and CNNs, show potential for predictive biocompatibility assessment.

Purpose of the Study:

• To compare the predictive performance of Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) models.
• To determine the most suitable AI approach for predicting scaffold biocompatibility.
• To utilize PrusaSlicer-generated scaffold designs for AI model training and validation.

Main Methods:

• Fifteen numerical design parameters were used to model scaffold biocompatibility with ANNs.
• Scaffold images were analyzed using CNNs.
• Models were trained on an 80/20 data split, with performance evaluated using accuracy, precision, recall, and F1-Scores.
• Experimental biocompatibility tests were conducted on five scaffolds for validation.

Main Results:

• The ANN model achieved perfect scores (1.0) in F1-Score, Precision, and Recall.
• The CNN model, with a batch size of 56, showed balanced performance with F1-Score (0.87), Precision (0.88), and Recall (0.9).
• The ANN model correctly predicted the biocompatibility of all five tested scaffold tissues, while the CNN model misclassified one sample.

Conclusions:

• ANN models demonstrate superior performance compared to CNN models for predicting scaffold biocompatibility using numerical design parameters.
• The study highlights the effectiveness of ANNs for structured data in bioprinting, improving prediction accuracy and efficiency.
• These findings can accelerate tissue engineering and personalized medicine by optimizing bioprinting processes and reducing costs.