When artificial intelligence (AI) is discussed in pathology, the focus almost inevitably lies on digital slide analysis. Algorithms that detect tumor regions, quantify biomarkers such as Ki-67, or support grading and classification dominate scientific publications, conference programs, and public debate. These developments are, without question, important and will continue to shape the future of diagnostic pathology, particularly as digital pathology becomes more widely adopted.
However, in everyday pathology practice, many of the most pressing challenges do not arise from diagnostic interpretation itself, but from the processes surrounding it. Rising case numbers, increasing diagnostic complexity, expanding regulatory requirements, and a growing shortage of qualified personnel place pathology laboratories under considerable pressure. In many institutions, these pressures are felt less at the microscope and more at the interfaces between laboratory, diagnostic work, documentation, communication, and administration.
Against this background, the question arises whether AI can contribute not only to diagnostic accuracy, but also to the organization, documentation, and long-term sustainability of pathological work. To answer it, we looked to the experience of one pathology institute that introduced a laboratory information system (LIS) with deeply integrated artificial intelligence for workflow optimization and structured medical documentation. The findings revealed that the non-diagnostic aspects deserve far more attention than they currently receive.
The institute in Bremen, Germany, is an anatomic pathology service handling approximately 50,000 cases annually. Its case mix includes complex surgical specimens from adjacent hospitals, as well as cases requiring state-of-the-art molecular pathology analyses.
The daily reality: bottlenecks beyond the microscope
Before integration of the new LIS system, the pathology institute operated with a conventional LIS and well-established workflows that were representative of many laboratories. Macroscopic descriptions and histopathological diagnoses were dictated by pathologists and subsequently transcribed by medical transcription staff. While digital transcription tools had been evaluated, the quality of the generated transcripts was considered insufficient for routine diagnostic use. Reports were then reviewed, corrected, validated, and released by pathologists.
This model had proven robust for many years and was supported by experienced teams. Nevertheless, its structural limitations became increasingly apparent as workload and complexity grew. Diagnostic information existed primarily as free text, tailored for human interpretation but poorly suited for reuse in downstream processes. For billing, cancer registry reporting, quality management, and internal statistics, relevant data had to be manually extracted from narrative reports and re-entered into different system components.
This fragmentation of information resulted in duplicated work, inconsistencies, and delays. One particularly visible consequence was the emergence of a backlog of dictated but not yet transcribed reports. Even with highly motivated and skilled staff, fluctuations in workload, staff availability, or case complexity could quickly lead to delays that affected turnaround times and, ultimately, clinical communication. Importantly, these challenges were not the result of insufficient commitment or expertise, but rather of structural limitations inherent to traditional documentation models.
Rethinking the role of the LIS
In many laboratories, the LIS is still perceived primarily as an administrative backbone: a system for case tracking, report storage, and billing. This perspective often limits the role of the LIS to documentation after the fact, rather than recognizing it as a central component of diagnostic workflows. AI, when present at all, is frequently added as a separate tool – most commonly for image analysis – rather than being embedded into the core of daily processes.
In this case study, the introduction of a new LIS with deeply integrated AI represented a deliberate departure from this paradigm. Instead of treating AI as an add-on, the system was designed around the idea that documentation, workflow coordination, and data reuse are central determinants of efficiency and resilience in pathology services.
From a user perspective, continuity was a key design principle. Pathologists continued to work primarily via dictation, both during gross examination and microscopic evaluation. However, the system no longer treated dictation merely as audio input to be converted into text. Instead, AI algorithms analyzed the spoken content in real time, identifying relevant medical information and assigning it to structured data fields. At all times, physicians were given the ability to review and correct content, and stayed in the role of author and signer.
During gross examination, the spoken information was not only structured and stored, but simultaneously used to generate a graphical representation of the number of specimens, cassettes, slides, and required staining protocols. Based on this information, the system automatically controlled connected hardware components, such as cassette printers, slide printers, and staining devices, without the need for manual case configuration or additional user interaction.
Macroscopic parameters, diagnostic statements, disease and staging codes, additional studies, and follow-up requests were automatically recognized and organized. All generated content was presented to the pathologist for review, correction, and final approval. At no point did the system perform autonomous diagnostic reasoning; responsibility for the report remained entirely with the physician.
From free text to structured information
This shift from purely free-text documentation to AI-supported structured information had profound implications. Diagnostic content was captured once, at the moment of interpretation, and became immediately available for multiple downstream purposes.
At the same time, the traditional separation between dictation, transcription, and subsequent physician review was effectively removed, as dictated content could be reviewed, corrected, and released directly by the reporting pathologist without the need for an intermediate transcription step.
Billing data no longer had to be reconstructed from narrative reports. Cancer registry notifications could be prepared directly from structured diagnostic elements. Internal statistics and workload analyses could be generated without additional manual effort. In effect, the LIS evolved from a passive repository for reports into a central hub for coordinated, data-driven processes across the institute.
One of the most tangible outcomes of this change was sustained reduction and eventual resolution of the documentation backlog. Reports were available for review and release without dependency on manual transcription workflows. Turnaround times from specimen accession to final report release decreased markedly, improving predictability and communication with clinical partners.
At the same time, the workload associated with manual transcription and documentation decreased. The regained time could then be redirected toward higher-value clinical, quality assurance, and organizational tasks.
Workflow integration and transparency
Beyond documentation, the system also enabled a more transparent and coordinated workflow. Dynamic work lists, real-time case status tracking, and clearly defined handover points made it easier for teams to prioritize tasks and identify bottlenecks early. This transparency supported not only efficiency, but also a shared understanding of workload distribution across professional groups.
Integration with laboratory hardware, such as cassette and slide printers, further reduced manual steps and potential sources of error. By linking documentation directly to laboratory processes, the system helped align diagnostic intent with technical execution.
These changes did not transform pathology overnight, nor were they free of challenges. Implementation required careful planning, structured training, and a period of adaptation. However, once teams became familiar with the system, the integrated AI features were used consistently and confidently in routine practice.
Acceptance: the human factor
One of the most important lessons emerging from this experience is that technological capability alone does not determine success. Acceptance by users is critical. In this setting, acceptance was facilitated by the fact that the system supported, rather than replaced, established working habits. Dictation remained central, but its role evolved from a documentation bottleneck into a powerful interface for structured data capture.
From the perspective of many users, AI operated largely in the background. It supported completeness, consistency, and efficiency without interfering with diagnostic reasoning. This helped address common concerns about loss of professional autonomy or overreliance on automated systems – concerns that frequently arise in discussions about AI in medicine and pathology in particular.
Beyond diagnostic AI
Experience from this implementation highlights a broader point that may be relevant for many pathology departments. While AI-based image analysis continues to evolve and attract significant attention, substantial and immediate gains can already be achieved by applying AI to workflow organization and documentation.
In an environment characterized by increasing workload and limited human resources, these applications may be just as important as advances in computational diagnostics. By reducing administrative burden and streamlining processes, AI can help pathologists focus on what matters most: diagnostic quality, clinical relevance, interdisciplinary communication, and teaching the next generation of specialists.
Lessons learned
Several lessons emerge from this experience. First, embedding AI deeply into core systems such as the LIS can yield tangible benefits without fundamentally altering diagnostic practice. Second, success depends not only on technology, but on thoughtful implementation, training, and sustained user engagement. Third, the impact of AI in pathology should be assessed not only in terms of diagnostic performance, but also in terms of workflow sustainability, staff satisfaction, and workforce resilience.
Looking ahead
The observations described here are based on real-world experience in one major anatomic pathology institute and should be interpreted accordingly. However, the underlying system architecture has also been implemented in pathology laboratories in Germany and the United States, suggesting that the described workflow principles are applicable across different healthcare settings. Nevertheless, they suggest that rethinking the role of AI in pathology – beyond image analysis – may offer valuable opportunities to address some of the most pressing challenges facing the field today.
As pathology continues to evolve, it may be worth broadening the conversation about AI. Sometimes, the greatest impact of artificial intelligence is not found inmaking diagnoses faster or more accurately, but in making everyday lab and pathology work sustainable for those who perform it.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
