Exploring transparency in pathological image analysis: A comprehensive review of explainable artificial intelligence (XAI) techniques

Image Analysis in Surgical Pathology

BackgroundThe Matlab software is a one of the most advanced development tool for application in engineering practice. From our point of view the most important is the image processing toolbox, offering many built-in functions, including mathematical morphology, and implementation of a many artificial neural networks as AI. It is very popular platform for creation of the specialized program for image analysis, also in pathology. Based on the latest version of Matlab Builder Java toolbox, it is possible to create the software, serving as a remote system for image analysis in pathology via internet communication. The internet platform can be realized based on Java Servlet Pages with Tomcat server as servlet container.MethodsIn presented software implementation we propose remote image analysis realized by Matlab algorithms. These algorithms can be compiled to executable jar file with the help of Matlab Builder Java toolbox. The Matlab function must be declared with the set of input data, output structure with numerical results and Matlab web figure. Any function prepared in that manner can be used as a Java function in Java Servlet Pages (JSP). The graphical user interface providing the input data and displaying the results (also in graphical form) must be implemented in JSP. Additionally the data storage to database can be implemented within algorithm written in Matlab with the help of Matlab Database Toolbox directly with the image processing. The complete JSP page can be run by Tomcat server.ResultsThe proposed tool for remote image analysis was tested on the Computerized Analysis of Medical Images (CAMI) software developed by author. The user provides image and case information (diagnosis, staining, image parameter etc.). When analysis is initialized, input data with image are sent to servlet on Tomcat. When analysis is done, client obtains the graphical results as an image with marked recognized cells and also the quantitative output. Additionally, the results are stored in a server database. The internet platform was tested on PC Intel Core2 Duo T9600 2.8GHz 4GB RAM server with 768x576 pixel size, 1.28Mb tiff format images reffering to meningioma tumour (x400, Ki-67/MIB-1). The time consumption was as following: at analysis by CAMI, locally on a server – 3.5 seconds, at remote analysis – 26 seconds, from which 22 seconds were used for data transfer via internet connection. At jpg format image (102 Kb) the consumption time was reduced to 14 seconds.ConclusionsThe results have confirmed that designed remote platform can be useful for pathology image analysis. The time consumption is depended mainly on the image size and speed of the internet connections. The presented implementation can be used for many types of analysis at different staining, tissue, morphometry approaches, etc. The significant problem is the implementation of the JSP page in the multithread form, that can be used parallelly by many users. The presented platform for image analysis in pathology can be especially useful for small laboratory without its own image analysis system.

Breast histopathological examination usually relies on experienced pathologists. In recent years, artificial intelligence technology has been applied to digital pathological images to help doctors diagnose breast tumor. This paper mainly introduces the research progress of artificial intelligence technology in pathological classification and histological grade of breast cancer. In view of the simplification of the pathological image analysis object of breast tissue, a multimodal breast cancer aided pathological diagnosis model is proposed. The proposed model combines the patient’s immunohistochemical diagnosis data and pathological image data with artificial intelligence to analyze and diagnose. To achieve a deeper evaluation of the patient’s condition on the basis of simple tumor screening, optimize the clinical treatment plan to make it more targeted. Discuss the main problems faced by artificial intelligence technology in the field of breast pathological image analysis, and prospect its development prospects.

This chapter thoroughly examines the critical role of artificial intelligence (AI) in drug discovery and development, covering its potential, methodologies, real-world applications, and the challenges it presents. It begins with a comprehensive introduction to AI and its subfields, including machine learning (ML), deep learning (DL), and natural language processing (NLP). The chapter then outlines various AI algorithms such as regression, support vector machines, and neural networks. It also explains approaches for optimizing and validating AI models, with a focus on metrics used for their quantitative assessment. Next, the chapter highlights the impact of AI across different stages of the drug discovery and development process, showcasing examples of its use in AI-driven drug discovery companies and their innovative platforms. Challenges such as limited data availability, ethical concerns, and integrating AI with traditional methods are discussed, along with potential solutions like data augmentation and explainable AI (XAI). It also explores regulatory perspectives, particularly from the United States Food and Drug Administration (FDA), illustrating the growing relationship between AI and regulatory science. The chapter concludes with a forward-looking view on AI's future in drug discovery. AI is revolutionizing the field by automating tasks such as image analysis in pathology and radiology, improving diagnostic accuracy, and reducing human error. In clinical trials, AI is used to optimize trial design, select appropriate patient groups, and monitor real-time data, leading to faster decision-making. AI also plays a key role in analyzing scientific literature, helping researchers stay current with new advancements.

Cognitive radios are expected to play a major role towards meeting the exploding traffic demand over wireless systems. A cognitive radio node senses the environment, analyzes the outdoor parameters, and then makes decisions for dynamic time-frequency-space resource allocation and management to improve the utilization of the radio spectrum. For efficient real-time process, the cognitive radio is usually combined with artificial intelligence and machine-learning techniques so that an adaptive and intelligent allocation is achieved. This paper firstly presents the cognitive radio networks, resources, objectives, constraints, and challenges. Then, it introduces artificial intelligence and machine-learning techniques and emphasizes the role of learning in cognitive radios. Then, a survey on the state-of-the-art of machine-learning techniques in cognitive radios is presented. The literature survey is organized based on different artificial intelligence techniques such as fuzzy logic, genetic algorithms, neural networks, game theory, reinforcement learning, support vector machine, case-based reasoning, entropy, Bayesian, Markov model, multi-agent systems, and artificial bee colony algorithm. This paper also discusses the cognitive radio implementation and the learning challenges foreseen in cognitive radio applications.

Deep learning (DL) is rapidly advancing in computational pathology, offering high diagnostic accuracy but often functioning as a "black box" with limited interpretability. This lack of transparency hinders its clinical adoption, emphasizing the need for quantitative explainable artificial intelligence (QXAI) methods. We propose a QXAI approach to objectively and quantitatively elucidate the reasoning behind DL model decisions in hepatocellular carcinoma (HCC) pathological image analysis. The proposed method utilizes clustering in the latent space of embeddings generated by a DL model to identify regions that contribute to the model's discrimination. Each cluster is then quantitatively characterized by morphometric features obtained through nuclear segmentation using HoverNet and key feature selection with LightGBM. Statistical analysis is performed to assess the importance of selected features, ensuring an interpretable relationship between morphological characteristics and classification outcomes. This approach enables the quantitative interpretation of which regions and features are critical for the model's decision-making, without sacrificing accuracy. Experiments on pathology images of hematoxylin-and-eosin-stained HCC tissue sections showed that the proposed method effectively identified key discriminatory regions and features, such as nuclear size, chromatin density, and shape irregularity. The clustering-based analysis provided structured insights into morphological patterns influencing classification, with explanations evaluated as clinically relevant and interpretable by a pathologist. Our QXAI framework enhances the interpretability of DL-based pathology analysis by linking morphological features to classification decisions. This fosters trust in DL models and facilitates their clinical integration.

Traditional analytic and time analysis approaches may not easily handle online real-time applications for large systems due to computational time requirements. In particular, the power systems being non-linear and time varying, application of traditional approaches to a power system for the purpose of identifying its parameters, controlling the operation to maintain stability and damping oscillations following disturbances is not suitable for online monitoring. They are more suitable for offline design and investigations. Advent of artificial intelligence (AI) techniques based on logic mathematics has encouraged power system engineers, planners and designers to employ these techniques with the goal of reducing computation time and designing fast algorithms that are adequate for power system online applications. Many AI and computational intelligence techniques, such as artificial neural network (ANN), fuzzy logic (FL), neuro-FL (NFL), particle swarm optimisation (PSO), genetic algorithms, exist. The basics of ANN, FL and NFL as well as the adaptive neuro-fuzzy control (ANFC) are presented in this chapter as they are used, in addition to the time analysis techniques, for some applications (e.g. power system stabilisers and static var compensators) to power systems in the subsequent chapters.

SI-ViT: Shuffle instance-based Vision Transformer for pancreatic cancer ROSE image classification

Digital pathology image analysis has become a new emerging research trend in the medical domain. AI methods have been shown their effectiveness on conventional vision applications. However, applying AI methods on pathology image analysis is far from easy. Many practical challenging issues arise including pathology image analysis under insufficient and inaccurate annotations, recognizing pathology images of different data distributions, and training AI models based on decentralized data sources. In this paper, we focus on discussing these challenging issues of AI approaches for pathology image analysis. A survey of relevant pathology applications will be also conducted. The research directions of these techniques for future development in pathology image analysis are also presented in this paper.

In recent years, the use of Artificial Intelligence (AI) techniques has been greatly increased. The term intelligence seems to be a must in a large number of European and International project calls. AI Techniques have been used in almost any domain. Application-oriented systems usually incorporate some kind of intelligence by using techniques stemming from intelligent search, knowledge representation, machine learning, knowledge discovery, intelligent agents, computational intelligence etc. The Workshop on Applications with Artificial Intelligence seeks for quality papers on computer applications that incorporate some kind of AI technique. The objective of the workshop was to bring together scientists, engineers and practitioners, who work on designing or developing applications that use intelligent techniques or work on intelligent techniques and apply them to application domains (like medicine, biology, education etc), to present and discuss their research works and exchange ideas in this book.

The analysis of pathological images, such as cell counting and nuclear morphological measurement, is an essential part in clinical histopathology researches. Due to the diversity of uncertain cell boundaries after staining, automated nuclei segmentation of Hematoxylin-Eosin (HE) stained pathological images remains challenging. Although better performances could be achieved than most of classic image processing methods do, manual labeling is still necessary in a majority of current machine learning based segmentation strategies, which restricts further improvements of efficiency and accuracy. Aiming at the requirements of stable and efficient high-throughput pathological image analysis, an automated Feature Global Delivery Connection Network (FGDC-net) is proposed for nuclei segmentation of HE stained images. Firstly, training sample patches and their corresponding asymmetric labels are automatically generated based on a Full Mixup strategy from RGB to HSV color space. Secondly, in order to add connections between adjacent layers and achieve the purpose of feature selection, FGDC module is designed by removing the jumping connections between codecs commonly used in UNet-based image segmentation networks, which learns the relationships between channels in each layer and pass information selectively. Finally, a dynamic training strategy based on mixed loss is used to increase the generalization capability of the model by flexible epochs. The proposed improvements were verified by the ablation experiments on multiple open databases and own clinical meningioma dataset. Experimental results on multiple datasets showed that FGDC-net could effectively improve the segmentation performances of HE stained pathological images without manual interventions, and provide valuable references for clinical pathological analysis.

The analysis of pathological images, such as cell counting and nuclear morphological measurement, is an essential part in clinical histopathology researches. Due to the diversity of uncertain cell boundaries after staining, automated nuclei segmentation of Hematoxylin-Eosin (HE) stained pathological images remains challenging. Although better performances could be achieved than most of classic image processing methods do, manual labeling is still necessary in a majority of current machine learning based segmentation strategies, which restricts further improvements of efficiency and accuracy. Aiming at the requirements of stable and efficient high-throughput pathological image analysis, an automated Feature Global Delivery Connection Network (FGDC-net) is proposed for nuclei segmentation of HE stained images. Firstly, training sample patches and their corresponding asymmetric labels are automatically generated based on a Full Mixup strategy from RGB to HSV color space. Secondly, in order to add connections between adjacent layers and achieve the purpose of feature selection, FGDC module is designed by removing the jumping connections between codecs commonly used in UNet-based image segmentation networks, which learns the relationships between channels in each layer and pass information selectively. Finally, a dynamic training strategy based on mixed loss is used to increase the generalization capability of the model by flexible epochs. The proposed improvements were verified by the ablation experiments on multiple open databases and own clinical meningioma dataset. Experimental results on multiple datasets showed that FGDC-net could effectively improve the segmentation performances of HE stained pathological images without manual interventions, and provide valuable references for clinical pathological analysis.

As the powerful performance of deep learning has been proven, many computer vision researchers have applied deep learning methods to their works as a breakthrough that could not be achieved with conventional computer vision algorithms. Particularly in pathological image analysis, deep learning plays an important role because some diagnosis requires a considerable cost or much time. In a recent, convolutional neural network (CNN)-based deep learning models have shown meaningful results in pathological image analysis, reducing time and cost. However, existing CNN-based segmentation models perform the same convolution operation for all channels of a feature map. It could be an inefficient operation according to information theory. We propose (Shannon) entropy-based convolutional module (ECM) for efficient convolutional operation in terms of a communication system. The fundamental coding manner of a communication system based on information theory is to allocate fewer bits for data showing the high probability of occurrence, and vice versa. Following up this coding manner, a feature is divided into dominant and recessive features according to the channel importance calculated from the channel attention module, and a heavy operation is conducted on the recessive feature and a light operation is conducted on the dominant feature. This operating manner can make a network perform efficient calculations and improve its performance. Furthermore, our proposed module is a portable unit, thus it can be a replacement of any convolution without modification of the whole architecture. To the best of our knowledge, our proposed module is the first trial to mimic the coding manner of information theory. The models equipped with our proposed module outperform the original models achieving 0.855 of F1 score and 0.832 of Jaccard score on colorectal cancer (CRC) image data-set.

This paper presents a snapshot view of the influence and direction of microcomputer technology for image analysis techniques in diagnostic pathology. Microcomputers have had considerable impact in bringing image analysis to wider application. Semi-automated tracing techniques are a simple means of providing objective data and assist in a wide range of diagnostic problems. From the common theme of reducing subjectivity in diagnostic assessment, an extensive body of research has accrued. Some studies have addressed the need for quality control for reliable, routine application. Video digitizer cards bring digital image analysis within the reach of laboratory budgets, providing powerful tools for investigation of a wide range of cellular and tissue features. The use of staining procedures compatible with quantitative evaluation has become equally important. As well as assisting scene segmentation, cytochemical and immunochemical staining techniques relate the data to biological processes. With the present state of the art, practical use of microcomputer based image analysis is impaired by limitations of information extraction and specimen throughput. Recent advances in colour video imaging provide an extra dimension in the analysis of multi-spectral stains. Improvements will also be felt with predictable increase in speed of microprocessors, and with single chip devices which deliver video rate processing. If the full potential of this hardware is realized, high-speed, routine analysis becomes feasible. In addition, a microcomputer imaging system can play host to companion functions, such as image archiving and transmission. With this outlook, the use of microcomputers for image analysis in diagnostic pathology is certain to increase.(ABSTRACT TRUNCATED AT 250 WORDS)

Computer-assisted methods for pathological image analysis can improve doctor's efficiency of image reading and diagnostic accuracy, effectively addressing the shortage of pathology diagnostic manpower. With the rapid development of artificial intelligence and digital pathology, deep learning technology has spurred a wealth of research in the field of histopathology. This article reviews the various applications of deep learning in digital pathological image analysis, such as pathological image segmentation, cancer auxiliary diagnosis, and cancer prognosis prediction, and discusses the challenges and solutions in its application. Furthermore, it predicts future trends in deep learning for pathological image analysis and proposes potential research directions.