Artificial intelligence and digital biomarker in precision pathology guiding immune therapy selection and precision oncology

Abstract Background The currently available immunotherapies already changed the strategy how many cancers are treated from first to last line. Understanding even the most complex heterogeneity in tumor tissue and mapping the spatial cartography of the tumor immunity allows the best and optimized selection of immune modulating agents to (re‐)activate the patient's immune system and direct it against the individual cancer in the most effective way. Recent Findings Primary cancer and metastases maintain a high degree of plasticity to escape any immune surveillance and continue to evolve depending on many intrinsic and extrinsic factors In the field of immune‐oncology (IO) immune modulating agents are recognized as practice changing therapeutic modalities. Recent studies have shown that an optimal and lasting efficacy of IO therapeutics depends on the understanding of the spatial communication network and functional context of immune and cancer cells within the tumor microenvironment. Artificial intelligence (AI) provides an insight into the immune‐cancer‐network through the visualization of very complex tumor and immune interactions in cancer tissue specimens and allows the computer‐assisted development and clinical validation of such digital biomarker. Conclusions The successful implementation of AI‐supported digital biomarker solutions guides the clinical selection of effective immune therapeutics based on the retrieval and visualization of spatial and contextual information from cancer tissue images and standardized data. As such, computational pathology (CP) turns into “precision pathology” delivering individual therapy response prediction. Precision Pathology does not only include digital and computational solutions but also high levels of standardized processes in the routine histopathology workflow and the use of mathematical tools to support clinical and diagnostic decisions as the basic principle of a “precision oncology”.


| INTRODUCTION
There is already a significant number of publications using AI and deep learning (DL) to identify novel diagnostic and prognostic biomarker signatures on tissue images of different cancer types. [1][2][3][4][5][6] Understanding the morphological and immunological complexity and plasticity of the cancer-related immune system in tissue is still one of the existing challenges in cancer immunotherapy. [7][8][9][10][11] The visualization of any contextual and spatial relationship of different immune, tumor and stromal cells, the communication network including humoral (extrinsic) and molecular (intrinsic) factors 5 will determine the selection of effective immune agents as a single compound or combination regimens. This becomes possible through the application of digital approaches, big data analysis and mathematical models, which go far beyond conventional techniques to answer important questions also in precision oncology. 12

| ARTIFICIAL INTELLIGENCE
With the advent of modern computing, many efforts are underway to replace, assist and augment human cognitive and analytic effort. It might not always be desirable, but it certainly allows addressing current objectives like the detection and readiness of complex immune biomarker. Such efforts are using AI attempting to create machine models for almost all aspects of human intelligence. 13 Some common definitions name AI as the heading for machine learning (ML), of which among others like deep learning (DL) and convolutional neural networks (CNN) are usually considered further subdisciplines. However, different and sometimes conflicting definitions exist, of which none are wrong or correct. Both utilize machine cognition technologies with different levels of supervision and guidance by human experts having domain knowledge. 14,15 The development of AI relies to some degree on already existing and conventional expert knowledge creating rule sets or algorithms that support clinical decision-making. AI's ability to describe current problems or anticipating problems of the future is also depending on the domain experts supporting such development. 16 2.1 | Machine learning AI will be pivotal for the future practice of pathology and oncology.
As mentioned above experts like pathologists and oncologists need to be involved in the development of AI-based decision support to ensure a professional digitization of the medical practice and the generation of clinically relevant algorithms through their already existing knowledge and clinical experience. ML usually applies stochastic methods to analyze data sets creating independent and sometimes novel rules. ML is considered an attempt to support human experience and expert knowledge. 17 A less ambitious goal is termed "narrow AI" which focuses on modeling presumably simpler tasks to support medical decision-making. If successful, it will allow the transition from narrow AI to broader AI. This comprises also different layers of advanced algebra and topology, 18 which describes the spatial relationship of immune cells and tumor cells and allows the functional cartography of tissue. [19][20][21] Mathematical and computer science techniques allow domain experts but also others to extract relevant data from large data sets. 22 Such algorithms can be trained or supervised by human experts or in this case by expert pathologists. ML can also assist experts in executing difficult and tedious tasks. An automated ML method will be able to consistently read multicolor immunohistochemistry or in-situ hybridization images always in the same reliable manner, 23,24 producing the identical result over and over again. Such a machine-assisted solution provides the basis for global comparability of even complex and larger data sets without otherwise non-acceptable inter-and intra-observer variability. 25,26 The field of AI-based solutions and algorithms in pathology provide an increasing number of diagnostic and therapeutic decision support tools. 27 The scanning and imaging of a whole (glass) slide has become a pivotal and prerequisite technology in histopathology that transfers conventional (analog) information into a high-quality digital image to apply existing algorithms or solutions for further and spatial analysis. 28 Only the precise, robust and reproducible diagnosis from tissue images will lead to an acceptance by pathologists with enough trust in such a disruptive technology. Already today, it is impressive how a computer can "read" a digitized image and "deliver" an accurate and quantitative interpretation, which goes beyond plain human eyeballing on a microscopic without computer assistance. Nevertheless, for the time being it is still necessary to confirm and validate any AI-assisted diagnosis through a highly skilled and well-trained pathologist concerning accuracy and plausibility. 29

| Visualization and explanation of data
Another important topic and prerequisite for the sustainable development of computational solutions in pathology as well as oncology is the use of curated data. Incorrect or inconsistent data will lead to incorrect conclusions and provide misleading decision paths. The cleaning and cleansing of data have a fundamental impact on the quality of such results. Therefore, data management and analytics needs to become an integrated part of the standard quality management and quality control throughout the entire workflow using computerassisted decision in clinical practice. 30 An intuitive visual representation of complex data to pathologists and oncologists allows the understanding of the used algorithms and extracted information explaining the rationale behind AI-based decision rules. The subject of topology as well as the cartography of the tissue microenvironment and its heterogeneity makes it now possible to further understand and visualize complex AI-based solutions of multidimensional data sets. 31 With the growing field of immune and combination therapies in precision oncology, a multitude of biomarker hypothesis will be integrated into topological networks, which will intuitively describe spatial relationships and relevant communication networks, 32 possibly leading to relevant treatment decisions. 33 AI tries to go beyond the "hidden secrets" or the "black box" nature of ML, which includes techniques such as convolutional neural networks, and DL. Those techniques allow an even deeper understanding and / or visualization of the complexity also of multidimensional features and cellular networks in heterogeneous tissue specimens to provide hypothesis or explanations of the results for pathologists' consumption and use. 34 Expert pathologists still verify the concordance between the AI-based decision rules and an already existing or accepted expert ground-truth. 35   The pivotal role of the pathologists is to master their responsibility from the bench to the bedside through their ability and growing experience and to implement and execute any type of (biomarker) assay robustly and sustainably, also in the routine diagnostic practice.
As digital and computational pathology advances, the role of pathologists will transform and extend including the documented management of stringent quality control measures of the laboratory workflow and the handling of biomarker analytics that generates more and more data to stratify patients. Increasingly more insights that are therapeutic important will also come from multidimensional tests including spatial transcriptomics and other context-driven information. 41

| Computational pathology
Recent advances in ML have accelerated computational pathology (CP) in medical research and clinical practice. Computational solutions will continue to support the diagnostic practice of pathology for yet well-defined and selected tasks but in a reliable, consistent, and standardized way. Pathologists who are faced with an increased and complex workload will appreciate computational support.
The potential of ML techniques in pathology ranges from computer aided support for tasks that are simple but tedious like counting colored dots but also the discovery of innovative biomarker signature. Basic applications with simple dichotomous decisions are the detection of lymph node metastases or counting the density of mitotic or Ki-67 positive tumor cells. CP is expected to increase the efficiency and precision in the entire tissue diagnostic workflow. First Tumor and immune heterogeneity heavily influence the biology of each tumor and its response to treatment, including therapy resistance and some uncertainty of the histomorphological diagnoses.
Genetic and epigenetic aberrations also influence the immune microenvironment and its plasticity and frequently vary from tumor entity to entity with or without previous therapy. Any failure of its identification may imply therapy relevant misinterpretations. 59-61

| Digital biomarker
Digital biomarker are generally defined as a combined softwarehardware solution to quantify measurable parameters that provide indications of a therapeutic response in a clinical environment. Digital biomarker also utilize data from different sources and measures to advance the understanding of a certain disease and guide the decision-making also in the diagnosis and treatment of cancer. 62 The idea of clinical immunotherapy is to (re)activate the immune system against uncontrolled tumor growth and spreading. 63 This is an especially difficult task in certain cancer types with all the existing and known immune escape mechanisms 64-66 that otherwise do not adequately respond to current strategies. [67][68][69] Some tumor entities are anyway hard to treat for various and sometimes obvious reasons. [70][71][72] Currently, there are no accepted biomarker signatures available for many immunotherapies. 73 One of the known diagnostic challenge is to understand, visualize and determine the biologically relevant spatial relationship and communication network in the tumor microenvironment and retrieve actionable and clinically relevant information.
Likewise, the analysis of multiple variables requires advanced technical tools and laboratory skills like high-resolution image acquisition and analysis and the application of ML-based algorithms to select patients for their best possible treatment option. Mathematical tools and AI-based solutions will help to gain confidence in technically assisted decision making along with necessary clinical trials and experience. This is exemplified in the description of tumor infiltrating lymphocytes 74 or the assessment of metastases in various tumors under immune therapy. 75

| PRECISION ONCOLOGY
The importance for advanced diagnostics to guide patient treatment decisions is growing fast. Table 1 describes the basic principles of AI and the use of ML in precision pathology as the foundation of precision oncology and their deliverables for best patient care. 76 Patients that lack access to advanced cancer pathology guided by computer-assisted diagnostic tools and expert decision boards are usually inadequately managed in their care. Expert pathologists are an increasingly scarce healthcare resource and therefore the "optimization" of their use especially important in times when their work is becoming more and more complex through AI-based tools. 77 Pathologists should engage with the AI development in pathology and its clinical implementation especially in precision oncology to assess its true value to the healthcare team.
Computer-assisted decisions are also based on data from the real world or selected cohorts or named register like TCGA and are further supported by modalities like "systems medicine" and "in-silico" modeling and simulation" approaches. AI will refine existing hypothesis of pathologists and immunologists to support diagnosis and therapy decisions.

| AI-supported immunotherapy
Many genetic factors explicitly modulate the immune microenvironment and can influence the selection of IO drugs or the combination of IO molecules with non-IO treatment regiments. [85][86][87][88][89][90][91] Similarly, cancer-associated fibroblasts (CAF) can also have a role in tumor progression and tissue remodeling secreting a wide range of humoral factors. 92 Moreover, CAFs can be the reason for developing a resistance to guideline-based therapies, as shown by Hirata et al. for BRAFinhibitor therapy in melanoma. 93 The understanding of the tumorwide heterogeneity and the contextual information concealed with T A B L E 1 Describes the basic principles of AI and precision pathology as the foundation of precision oncology. Their deliverables and effects will lead to a deeper understanding of the tumor biology and explaining even complex cancer networks that will better guide therapy selection for individual patients  [100][101][102]105 Further studies emphasized the relevance of cellular components in the immune system besides (epi)genetic biomarker. Both and in particular their spatial co-existence have a significant prognostic value and need be included in therapeutic considerations. [103][104][105][106] Table 2 lists examples of AI-assisted and digital biomarker in precision pathology and oncology with a special emphasis on immunotherapy.
The future of precision cancer care will not only include the use of AI-based algorithms in precision pathology and the diligent use of digital biomarker, but also major efforts to detect cancer earlier and with greater accuracy. The availability of larger data sets and a wider range of information from many sources (e.g. liquid biopsies, other imaging techniques) will help to identify the most effective treatments in a particular cancer and individual patient. Precision oncology will stratify many patients towards the most optimal cancer care right from the beginning which might be more effective, less costly and more likely to result in better overall outcomes.

| CONCLUSION
Precision pathology will be the foundation and driver for precision oncology and immune therapies extending treatment regiments including oligo-metastatic diseases or targeting the tumor microenvironment independent of the origin of the primary cancer. The basic principle is briefly summarized in Figure 1. Besides managing the "tumor data business", also the technical and laboratory pathology workflow will change drastically leaving glass slides, conventional stains and eventually the light microscope behind and embracing 3D-imaging including augmented and multiplex visualization techniques supported by AI. The implementation and execution of precision oncology including immune and combination therapies will be part of the medicine of the 21st century and pathologists will (co)lead such efforts embracing precision pathology.

ACKNOWLEDGMENTS
There was no external financial support of this work. We are grateful to all colleagues at the Institute of Pathology and Molecular Diagnostics as well as the Institute for Digital Medicine for the continuous discussions and advice.

T A B L E 2
Lists examples of AI-assisted decision support in precision pathology and those tools that have an even broader clinical impact in precision (immune) oncology. Digitale biomarker are also AI based but combine more digital and quantifiable characteristics from different sources that have a relevant impact on the clinical practice, here especially in the selection of immune therapies