Blog 6: Deploying an AI Diagnostic

The journey of developing an AI diagnostic tool is long and intricate. Rigorous development, validation, and regulatory approval are required before clinical use. The final step, deployment, brings this innovative tool into the hands of pathologists, integrating it seamlessly into their workflows. This integration requires meticulous planning to ensure ease of use and reliability.

In this final blog of our series, we explore the deployment of an AI diagnostic into a pathology lab, highlighting successful integration strategies and the transformative benefits these tools can bring to pathology.

Digital readiness and seamless workflow integration

Integrating AI diagnostics is feasible in laboratories that have already embraced digitization. Digital pathology labs equipped with whole slide imaging (WSI) systems and information management systems (IMS) are well-positioned to incorporate AI tools seamlessly. WSI systems allow for the digitization of tissue slides prepared for microscopic analysis, producing high-resolution images that can be viewed, shared, and analyzed on computers. This digital readiness ensures that the AI diagnostic can be integrated without disrupting existing workflows.

Once the laboratory is digitally ready, pathologists can integrate the AI diagnostic into their existing workflow. It must work harmoniously with a lab’s informational systems, such as the IMS or Laboratory Information System (LIS). This integration allows AI annotations, heatmaps, or scores to appear alongside slides for pathologists to review.

With the variety of scanners and IMS software available today, zero-click integrations are available. These integrations enable the AI to operate in the background, automatically analyzing slides, generating results, and integrating findings directly into the IMS and LIS. This allows pathologists to access AI insights without additional steps and enhances their workflow efficiency.

For the AI diagnostic to be fully adopted, pathologists must understand its value and trust its results. This trust is built through transparency in the development process, rigorous validation, and real-world testing.

Defining value and building trust

The value of an AI diagnostic is multifaceted, addressing critical clinical, operational, and financial needs:

Clinical impact: enhancing diagnostic precision, reducing errors, detecting disease earlier, providing new methods of analysis, and improving patient outcomes.

Workflow efficiency: reducing turnaround time, increasing pathologist productivity, conserving resources, and prioritizing cases.

Cost-effectiveness: saving costs, avoiding downstream expenses and errors, and increasing scalability.

Accessibility: expanding access to smaller or underserved labs and broadening global access.

Research and innovation: providing new insights and accelerating clinical trials.

‍

This value is quantified during the validation phase. For example, studies found that by using Owkin’s MSIntuit CRC diagnostic for screening, almost 50% of colorectal cancer patients can be ruled out from reflex IHC/PCR for MMR/MSI testing, triaging the screening for fewer cases more likely to have MSI. For breast cancer, RlapsRisk BC has the potential to identify women who can avoid toxic chemotherapy that would not be beneficial to them, in their treatment plan.

Each AI diagnostic provides different insights into patient samples. It is up to the pathologist to integrate this new information with their other views of the patient and make informed decisions.

As this technology becomes more prevalent, its level of impact to augment research, accelerate clinical trials, and ultimately benefit patients will deepen. The extensive validation and real-world testing ensure that AI diagnostics are reliable and effective before they are used in clinical practice.

Transformative potential

Figure 1: Pathologists’ opinions on the use of AI for Ki-67 assessment in breast cancer.¹

The integration of AI diagnostics can significantly alleviate the stress and burnout experienced by pathologists. A recent survey revealed that 55% of pathologists and oncologists face high stress or burnout due to increasing workloads and the complexity of diagnoses. AI tools are expected to cut waiting times for tests and help with staffing challenges, as highlighted by 77% of surveyed professionals.

AI has the potential to revolutionize pathology by streamlining workflows, automating routine tasks, and providing new insights into patient outcomes and treatment responses. With growing trust in AI diagnostic tools, pathologists are increasingly confident in their use, with 82% expressing some level of confidence and 70% believing their patients are somewhat/or completely comfortable with AI.

Figure 2: State of the Nation: Opportunities and Challenges for Digital Pathology and AI Solutions.²

Conclusion

The deployment of AI diagnostics in pathology represents a significant leap forward in enhancing diagnostic accuracy, streamlining workflows, and improving patient outcomes. As we conclude this blog series, we hope we have demystified the development process and highlighted the transformative potential of AI pathology. The journey from data to deployment is complex, but the benefits are profound. As these processes become more refined and standardized, the availability of AI diagnostics for pathologists will grow. Ultimately, pathologists will be responsible for embracing these tools, interpreting AI-generated insights, and making informed decisions to provide the best possible care for their patients.

Stay tuned for more insights and innovations from Owkin Diagnostics as we continue to push the boundaries of digital pathology and AI solutions.

‍

¹Amanda Dy et al. 2024. AI improves accuracy, agreement and efficiency of pathologists for Ki67 assessments in breast cancer. Sci Rep. 2024 Jan 13;14(1):1283. doi: 10.1038/s41598-024-51723-2

² Owkin’s State of the Nation report: Opportunities and Challenges for Digital Pathology and AI Solutions