Artificial Intelligence in Digital Pathology Image Analysis

Thanks to the development and deployment of whole-slide imaging technology in pathology, glass slides previously observed under a traditional microscope are now scanned and converted to digital images, which are more beneficial for remote access, portability, and ease of sharing to facilitate telepathology. More importantly, digitization of glass slides paves the way towards the wide use of artificial intelligence (AI) tools including machine/deep learning algorithms, resulting in improved diagnostic accuracy. In the past decade, a large number of studies have demonstrated the remarkable success of AI, particularly deep learning, in digital pathology, such as tumor region identification, metastasis detection, and patient prognosis.

Differing from handcrafted feature-based approaches that take advantage of domain knowledge to delineate specific morphological measurements (e.g., nuclei shape and size and tissue texture) in the images as features for training, deep learning is a paradigm of feature learning entirely driven by the image data and/or labels. Herein, the use of deep learning in pathological diagnosis can not only handle increased workloads and expertise shortages but also obviate subjective diagnosis from pathologists. Yet there remain many scientific and technological challenges associated with the efficiency of deep learning algorithms for use in clinical practice. For example, deep learning requires a sufficient amount of training data for generalization and suffers from a lack of feature interpretability. The overarching goal of this special issue is to highlight novel research accomplishments and directions, related to advanced AI methodology development and applications in digital pathology. Prospective authors are invited to submit Original Research as well as Review articles on topics including but not limited to the following:

• New machine/deep learning architectures for tumor region identification/image segmentation;


• New machine/deep learning architectures for downstream applications, such as histological subtyping, disease grading, and precision medicine;


• Multi-modal machine/deep learning strategies for fusion of multi-modal imaging data (e.g., pathological and radiological images) and/or genomic and proteomic data to improve diagnostic prediction;


• Interpretability of deep learning algorithms in digital pathology;


• Unsupervised or weakly-supervised machine/deep learning algorithms in digital pathology, especially where few or only weakly annotated data are available;


• Privacy-preserving distributed machine/deep learning;


• Integration of handcrafted features and deep learning features to yield superior performance as well as to detect biomarkers;


• Integration of pathological images and sequencing data in predicting disease-associated biomarkers or disease diagnosis and prognosis.

We would like to acknowledge the contribution of Prof. Binsheng He and Dr. Jiangqiang Sun to the development of this Research Topic as Research Topic Coordinators.