This website uses cookies to improve your experience.

Cookie policy

Towards optimisation in surgical pathology – the potential of Artificial intelligence

Authors: Irene A. Spiridon, MD, PhD1,2; Barbara Seeliger, MD, PhD1,3,4,5

1) Institute of Image-Guided Surgery, IHU-Strasbourg, Strasbourg, France
2) “Grigore T. Popa” University of Medicine and Pharmacy, Department of Morpho-functional Sciences I – Pathology, Iasi, Romania
3) Department of Digestive and Endocrine Surgery, University Hospitals of Strasbourg, Strasbourg, France
4) ICube, UMR 7357 CNRS, University of Strasbourg, Strasbourg, France
5) IRCAD, Research Institute Against Digestive Cancer, Strasbourg, France

Funding: This work was supported by French state funds managed within the “Plan Investissements d’Avenir” and by the ANR (reference ANR-10-IAHU-02).
Towards optimisation in surgical pathology – the potential of Artificial intelligence

The ever-evolving environment of surgery continues to rely on pathology examination as a diagnostic pillar and fundamental guide for pre- and postoperative patient management, particularly in the field of multidisciplinary cancer treatment1. For an accurate diagnosis, a comprehensive pathology report requires a thorough assessment of the surgical specimen, which routinely encompasses gross examination, microscopic evaluation and additional testing in order to provide the information needed for patient management 1. In the current era of personalised medicine, disease entities and classifications are continuously amended in response to emerging molecular data, which are incorporated for novel therapies1, 2.

Histopathology and surgery – expectations and timing

In the preoperative phase, histopathological information is required for multidisciplinary decision-making regarding the indication and timing of surgery within oncological treatment pathways. Intraoperatively, quick diagnostics allow the surgical strategy to be tailored (suspected malignancy, resection margin assessment, sentinel node strategy versus extended lymphadenectomy, synchronous versus skip lesions,) or provide fresh tissue for additional studies (molecular, electron microscopy, microbiology). Postoperatively, comprehensive specimen analysis orientates towards oncological surveillance versus adjuvant treatment, recommendations for genetic analyses and other profiling techniques for tailored patient management2. To that end, numerous tasks stand to benefit from digitalisation and optimisation assisted by artificial intelligence (AI), especially in medical centres with high workloads3.

A full pathology report is based on a variety of tasks that fall under the responsibility of technicians, residents, and consultants4. Quality control of surgical biopsies is imperative to identify and correct potential errors such as a misidentified sample, inappropriate container, insufficient fixation, contamination, tissue adequacy5 and, for cytology samples, staining quality and determination of cellularity6. To minimise such errors, an active interdisciplinary exchange is mandatory to agree on specialty-specific terminology and technical considerations.

Prioritisation and automatic requests for complementary analyses can optimise the workflow and shorten the turnaround time to final diagnosis4. An optimised workflow in surgical pathology makes it possible to meet the challenge of keeping up with evolving guidelines and management protocols under time pressure and in sometimes understaffed departments. Time-sensitive procedures such as intraoperative frozen-section examination or biopsy examination in oncological settings have a strong impact on surgical decision-making and overall patient management7. An illustration of standard and digitally enhanced workflows is provided in Figure 1.

Figure 1

Intraoperative synergism

Intraoperative pathology consultation is a mainstay of surgical management in many subspecialities and requires speed and optimal communication between surgeons and pathologists. Frozen section examination is incorporated into surgical routine for lymph node status and surgical margin assessment. Nevertheless, in addition to the complexity of cancer biology and its impact on morphological assessment, some challenges remain.

A number of emerging real-time techniques such as radio-frequency spectroscopy, bioimpedance spectroscopy, ultraviolet photoacoustic microscopy, optical coherence tomography, chemistry‐based fluorescent probes, hyperspectral imaging and others have been proposed with the aim of replacing intraoperative pathological consultations8. Despite promising first results, most of these novel methods do not yet benefit from broad clinical validation and are not yet available outside of clinical studies, or even affordable for most surgical units.

Novel intraoperative imaging technologies in conjunction with computer-assisted surgery aim at real-time enhancement of the surgeons’ eye, including a microscopic view (Figure 2). The ideal solution is in high resolution and in real time, with little or no impact on surgical workflow, while being safe for the patient and easy to use9. Non-invasive contrast-free optical imaging technology such as hyperspectral imaging is promising as it acquires spatial and spectroscopic data in real time10. Since the light interactions depend on biochemical tissue properties, hyperspectral imaging provides quantitative information on parameters like oxygen saturation and water content10, and the different reflectance curves allow discrimination between anatomical structures as well as between normal and neoplastic tissue. This information, in conjunction with machine learning approaches, could automate intraoperative tissue recognition9.

Figure 2.

Among the technologies that allow access to the microscopic optical domain in real time, confocal laser endomicroscopy (CLE) deserves special mention. It is a probe-based high resolution microscopic imaging system with a penetration depth of ~250μm, providing “virtual biopsies”11. However, it involves injection of a fluorophore, and image interpretation is operator-dependent requiring dedicated training9. Machine learning can be used to generate automated detection of specific tissues and aid in the decision-making process, as seen in ex vivo studies on neoplastic skin disease, with its use in in vivo settings being highly anticipated12.

Since its first description in 1991, non-invasive optical coherence tomography (OCT) has been used in ophthalmology and cardiology 8. Its capacity to generate instant images of tissue using reflected light without the need for specific staining or sample processing makes it a suitable method for both in vivo and ex vivo “optical biopsies”. Its use expanded across other surgical domains such as thoracic (bronchial tumour detection), abdominal (hepato-biliary and pancreatic, premalignant colorectal lesions), plastic surgery (malignant melanoma and basal cell carcinoma) 13. Technological progress improved image resolution (full-field OCT) and added functional parameters such as intracellular activity (dynamic full-field OCT), but also led to adaptation to flexible endoscopic and ultrasound exploration8, 9, 13, 14. OCT imaging shows great promise in intraoperative tumour detection, surgical resection margin evaluation and lymph node status, but requires further validation in large-scale clinical studies.

Given the lack of widespread availability and the challenges of increased costs, staff training and equipment maintenance for modern image-guided intraoperative techniques, frozen section examination remains the gold standard for intraoperative tissue assessment. Contemporary approaches to time-sensitive intraoperative consultation, particularly for remote or understaffed medical centres, imply the use of telepathology, especially since the COVID pandemic facilitated the adoption of institutional teleconference systems. A recent study to validate telepathology has shown a promising concordance of 96.4% between digital slide evaluation and the classical light-microscopy examination. Such approaches pave the way for another version of the “intraoperative pathology laboratory”15. Developments in deep-learning algorithms have shown both noteworthy support of intraoperative pathology consultation and improvement in the accuracy and speed of frozen section examination16.

Post-surgical assessment

Radiology paved the way for computer-assisted diagnostic tools17, and digital pathology is following in its footsteps. Interdisciplinary collaboration can benefit from computer-assisted healthcare solutions by aiding in the orientation and macroscopic description of surgical specimens via automated localization, recognition, tracking of objects/areas of interest and topographic mapping of samples, with proven results in both minimally invasive and open approaches18, 19. To this effect, whole-slide imaging (WSI) of tissue sections mounted on glass slides is demonstrating its usefulness17. AI systems can be assimilated into the pathology laboratory workflow to assist every step leading up to a complete pathology report, resulting in the emergence of a new field called computational pathology (cPath)2, 4. The main features of this visionary healthcare-integrated approach include accuracy, productivity and discovery2.

Recent years have seen the progression of the infrastructure for this endeavour and WSI can now benefit from more ergonomic solutions in scanning, storage of digital data, and various image management and processing technologies involved in routine diagnosis5. Following the four pillars of pathology-orientated AI assistance (detection, classification, segmentation, and quantification)2, deep-learning techniques benefit the analysis of surgical specimens by shortening decision-making times and allowing for diagnosis optimisation5. Algorithm-assisted pathologists showed a significantly improved quality of work output in accuracy and sensitivity as compared to unassisted pathologists5 (Figure 3).

Figure 3.

CPath applications for microscopy-orientated techniques include detection of tumour type, invasion patterns, metastatic dissemination, lymph node status, and spatial assessment of tumour heterogeneity including inflammatory infiltrate, vascularization patterns, histological grading and other types of scoring20. Routine examination frequently implies several of these parameters. Deep learning architectures are utilised for training and applications in pathology in order to reduce human error rates and improve workflow5. Most of pathology AI-driven research is in the field of cancer, with promising emerging data on how computer-assisted detection can increase diagnostic accuracy in the case of gynaecological tumours21, bringing clarity to challenging issues such as histological grading of prostate carcinoma22 or providing better outcome estimates for gastrointestinal malignancies23. Moreover, these concepts can be integrated into technological solutions for non-neoplastic pathologies. As recently demonstrated, AI techniques applied to histopathological samples can contribute to predictive models with high accuracy in anticipating disease activity in ulcerative colitis24.

The “tumour tissue journey”, as coined by some experts, is not a fixed path, but an evolving one25. Advances in research translated into medical practice bring new personalised decision-making based on patient-specific information provided by histopathology5. This comprises immunohistochemical evaluation of samples, a process that already benefits from technological advances through automated cell counting systems, and is increasingly optimised through deep learning approaches to provide accurate and rapid results2.

In addition to histopathological diagnosis, the molecular profiling of cancers has proven indispensable in modern-day medicine for improving patient outcomes, and can be expanded to the generation of novel molecular biomarkers, tailoring of targeted therapies, and broader topics like intrafamilial risks and transmission of diseases evaluation. Most facilities in which molecular testing is available already benefit from advanced digital infrastructure to perform genomic analyses on formalin fixed paraffin embedded (FFPE) samples to characterise tumours and evaluate therapeutic schemes (including next-generation sequencing)26. The samples are defined by tumour purity given as a percentage and are highly variable in terms of tumour size and laboratory equipment. This is another area in which WSI-analysis and computational assistance reduce error rate and timelapse27. Furthermore, the convolutional neural networks employed by computer-assisted approaches can also be trained automatically to identify tumour varieties and accordingly order complementary tests. Optimised data collection and integration leads to a comprehensive diagnosis, a process currently demonstrated in the field of central nervous system tumours5, 28.

After diagnosis, the integration of the pathology report into clinical management is done by the primary physician. Remote locations or understaffed laboratories can benefit from securely sharing WSI for diagnostic purposes, and expert opinion in difficult cases is more readily available through telepathology23. Furthermore, AI algorithms can be employed as decision support tools for multidisciplinary tumour boards, ultimately serving as an orientation tool in the quest for personalised cancer treatment 26, 29.

Special considerations

The introduction of WSI as part of the digitalisation movement in medicine has sparked numerous controversies, underlining the need for standardisation5. Accordingly, the College of American Pathologists (CAP) published the first guidelines for WSI systems validation in 2013, providing a basis for regulations that now surround the use of AI in anatomical pathology15. Professional organisations, such as the Digital Pathology Association, the Association of Pathology Informatics or the European Society for Digital and Integrative Pathology, rely on interdisciplinary communication to close the gap between different medical specialities in order to provide the best support for optimisation of patient care in accordance to the latest state of the art4. The collaboration between surgery and pathology aims not only to provide the best standard of care, but also to adjust currently existing procedures and to improve perioperative patient prognostication. Integrated approaches are crucial for further advances, such as the ones enabled using WSI and telepathology, allowing for expert consultation and unprecedented teaching opportunities2, 15.

While complete workflow digitalisation requires a significant update of equipment, a stepwise approach can prove beneficial in integrating AI-based technology and optimise certain aspects of daily practice, particularly in relation to surgical departments28. A recent study evaluating the opinion of an international expert panel points towards the certainty that specific pathology AI applications will be included in diagnostic routine by 20304.

The use of AI in pathology has already facilitated the emergence of digital biomarkers, defined as indicators of disease progression (risk association, disease monitoring, response to therapy, prognosis), and their inclusion into medical practice is highly anticipated 2.

Telepresence involves both remote mentoring and remote manipulation, and the establishment of a global connectivity infrastructure promotes the democratisation of expertise. Remote collaboration has the potential to support telepathology not only through telediagnosis, but also through remote control of a robotic system. The ground-breaking Lindbergh Operation, in which a laparoscopic cholecystectomy was performed remotely across a transatlantic distance, has already demonstrated the safety and feasibility of telesurgery30. As digitised medicine is permeating into routine practice, the avenue of computational pathology integration into surgical robotics is only starting to unveil its potential.


Artificial Intelligence-based solutions applied to pathology have already penetrated the medical market and are increasingly used to optimise workflow, while also facilitating surgically oriented requests5. With the large amount of digitised data comes the need for structuring and standardisation centred around the human expertise. The current outlook of pathology laboratories is very likely to change in the coming years, much like the concept of the “operating room of the future”, while acknowledging the indispensable role of the physician for optimal patient care.

Figure 1: Standard versus digitally enhanced workflow.

Standard clinical workflow (left) with specimen transport to the pathology laboratory accompanied by a handwritten note, manual tissue processing, and communication of the frozen section or final report. Digitally enhanced workflow (right) with automatised specimen transport to the pathology laboratory, accompanied by a digital note ± an intraoperative image, AI-assisted specimen analysis, automatised requests for complementary analyses, and detailed digital report generation (computational pathology).

Figure 2: The ideal interconnected operating room integrating advanced intraoperative imaging technologies for precision surgery in a digital environment.

Fluorescence imaging is already available within the most widely used robotic systems and is based on injection of a fluorophore (indocyanine green) to enhance tissue contrast and visualise perfusion. Confocal laser endomicroscopy provides a magnification that enables to visualise cellular movement in vivo, based on intravenous injection of a fluorescent dye (fluorescein).

Hyperspectral imaging integrates spatial and spectral information without an exogenous dye to visualise tissue-specific and perfusion-related characteristics.

Full-field optical coherence tomography enables a high-resolution view like an “instant microscopic examination” without requiring any specific tissuepreparation or staining.

Figure 3: Whole-slide imaging (WSI) and digital analysis.

(A) Through scanning, a conventional glass slide is transformed into a digital image. (B) The concept of WSI is based on multiple captures at different magnifications, generating a pyramidal-like structure. (C) Details from any area of the slide can be accessed through magnification, without affecting the quality of the image, and most systems also permit to digitally zoom to a certain extent. (D) Haematoxylin-eosin (HE) staining of a gastrointestinal stromal tumour displaying epithelioid morphology and congested vessels of the stroma. This colouring technique allows the clear differentiation between the nuclei of the cells (purple) and the cytoplasm and extracellular matrix (pink). (E) Owing to this chromatic distinction, automated image analysis using various available software solutions can be employed to replace certain tasks such as cell counting – the image shows the outlines of the nuclei and the cell contour in red (QuPath).


Part of the charitable activity of the Society, BJS Academy is an online educational resource for current and future surgeons.

The Academy is comprised of five distinct sections: Continuing surgical education, Young BJS, Cutting edge, Scientific surgery and Surgical news. Although the majority of this is open access, additional content is available to BJS subscribers and strategic partners.

Discover the Academy
Surgeon Training & Surgeons in Surgery