Abstract
Artificial intelligence (AI) refers to a diverse set of technologies that are designed to iteratively solve problems, optimize performance, and perform complex tasks otherwise associated with human intelligence—often by utilizing machine learning (ML) and other computational methods to “learn” underlying patterns within big data. While the origins of AI as an academic discipline can be traced back as early as 1956, modern applications of AI and ML are highly interdisciplinary and have now been integrated with numerous fields of science and engineering. 1 In the study of tissue engineering (TE) and tissue biology, AI methods have supported the traditional scientific process by several means. These include (1) identification of complex, nonlinear relationships between biological variables, (2) prediction of biomedical outcomes based on multimodal data, (3) accelerated discovery through experimental optimization, and more. 2 In each of these areas, the adoption of AI-based technologies has rapidly grown as these tools have become increasingly accessible to scientists of all backgrounds.
Excitingly, tissue engineers and biologists have now applied AI and ML methods for a wide array of applications that includes clinical grading and prognosis, cell and tissue morphological analysis, biomaterial discovery and optimization, and many other scientific areas. In this Special Issue, we highlight the breadth and depth of interdisciplinary research at the convergence of AI, tissue engineering, and tissue biology, an area which has recently exhibited exponential growth in both publications and public interest. 2
In the first paper of this Special Issue (“Deep Learning Augmented Osteoarthritis Grading Standardization”), Nagarajan et al. apply deep learning algorithms to automate image classification for the grading of cartilage histology. 3 Traditionally, this is a labor-intensive task that can also present significant interobserver variation, complicating efforts to perform quantitative and/or predictive assessment. By standardizing the classification of histological images using neural networks, they are able to achieve strong ROC-AUC values from 0.89 to 0.99 and a validation accuracy near 84% in grading knee osteoarthritis, producing algorithmic scores in accordance with gold standard classifications from medical experts. This publication demonstrates the effective application of AI for this challenging clinical task.
The next paper of this Special Issue (“Hematoxylin and Eosin Architecture Uncovers Clinically Divergent Niches in Pancreatic Cancer”) by Guo et al. quantifies intratumoral extracellular matrix (ECM) architecture for pancreatic cancer prognosis. 4 The authors extract 147 fiber features from tumor biopsies stained with hematoxylin and eosin (H&E), a routine histological stain used in existing clinical workflows. From this high-dimensional feature set, they summarize patient-specific profiles of ECM architecture that correlate with clinical outcomes such as survival time and disease recurrence. Further, the authors identify in situ cellular niches that colocalize with regions of differential tissue architecture, highlighting outcome-negative features such as B cell infiltration, pericyte enrichment, and inflammatory protein expression. This paper illustrates how AI can be utilized to identify complex spatial-biological motifs that correlate with clinical outcomes more effectively than traditional biomarkers.
In the following paper (“Revealing Early Spatial Patterns of Cellular Responsivity in Fiber-Reinforced Microenvironments”), Pucha et al. utilize ML to identify patterns of mesenchymal stromal cell (MSC) response to fiber reinforcement within TE scaffolds. 5 Using fibrin gels supplemented with polyglycolide-cocaprolactone (PGCL) fibers, the authors demonstrate that MSCs exhibited 3 “clusters” or subtypes of mechanoresponsivity based on 23 quantified features of cellular morphology and mechanoresponsive gene expression. They apply agglomerative hierarchical clustering to resolve the heterogeneity of MSC response, identifying enrichment of mechanoresponsive MSC subtypes closer to PGCL fibers. Additionally, the authors utilize their cluster-based model to assess how microenvironmental factors such as gel stiffness and remodeling impact cellular morpho-mechanoresponse. This work demonstrates how AI techniques can help resolve complex, nonlinear relationships between the biomechanical features of engineered tissue.
Next, Chechekhina et al. review the applications of code-free deep learning tools for the analysis of microscopy images (“Code-Free Machine Learning Solutions for Microscopy Image Processing: Deep Learning”). 6 The authors highlight image-based algorithms that have gained traction for the study of biology, particularly tools that are catered towards users without formal programming expertise. In addition to describing the fundamental principles of deep learning for image processing, the authors provide a detailed overview of different application areas, including cell segmentation, object detection, image translation, and more. This timely review provides tissue engineers and biologists with a highly accessible guide for using deep learning tools in biological image analysis.
Viet et al., in the next paper (“Artificial Intelligence Applications in Oral Cancer and Oral Dysplasia”), review ongoing efforts to apply AI for the identification of patients at highest risk for developing oral squamous cell carcinoma (OSCC). 7 The authors describe how ML- and deep learning-based digital pathology is currently being pursued to assess oral epithelial dysplasia and OSCC outcomes. Additionally, they highlight the integration of these AI-based methods with emerging biological data modalities in oral cancer biology such as multiplexed immunohistochemistry and epigenomics, in order to identify both prognostic and therapeutic targets. Overall, this review showcases the translational potential of AI to augment established histopathological predictors of OSCC progression and to uncover previously undescribed biological patterns present in omic-scale data.
In the following paper (“MyoFInDer: an AI-based tool for Myotube Fusion Index Determination”), Weisrock et al. describe a novel Python-based program for automated computation of the fusion index, a key indicator used for quantifying the differentiation of myoblast populations. 8 This AI-based image segmentation model also determines the total nuclei count and percentage of stained area, in addition to allowing for manual verification and correction. The authors show that the MyoFInDer tool strongly correlates with the results of manual counting and also minimizes interoperator variability compared with existing computational tools. They provide this program as a free and open-source project, supporting the potential of AI-based methods to accelerate in vitro analyses and protocols that are highly relevant to the tissue engineering community.
In the final article (“Mapping Biomaterial Complexity by Machine Learning”), Ahmed et al. review the applications of ML in uncovering novel structure-function relationships relevant to biomaterial design. 9 The authors describe how ML can be combined with high-throughput experimentation to uncover complex associations between multiple chemical, physical, and biological properties that drive biomaterial function, including strategies by which ML can be used to address the “curse of dimensionality” associated with large, multifactorial datasets. They highlight ongoing developments in ML for a number of biomaterial applications, including tissue engineering, drug delivery, antifouling, protein stabilization, and other areas. Additionally, the authors discuss data-mining approaches that may help reduce experimental load and accelerate biomaterial discovery by identifying quantitative patterns within published biomaterials data. This article comprehensively outlines how AI and ML can be used to model and/or optimize the complex, heterogeneous features that determine biomaterial function in the context of tissue engineering and biology.
Ultimately, this Special Issue highlights the encouraging scientific and clinical potential now arising at the intersection of AI, tissue engineering, and tissue biology. As AI continues to grow in scientific adoption, the Guest Editors hope that this issue will spark broad interest in the readership for the diverse, highly creative applications of AI in both engineering and analysis of biological tissue.