AI-driven abnormality detection via cloud-based software-as-a-service to boost the productivity of toxicologic pathology
Toxicologic pathologists meticulously review thousands of microscope slides of nonclinical studies in search of abnormalities related to drug treatment. Currently, digital tools offer help for slide review but this process is less efficient than using conventional microscopes. Patholytix Preclinical(opens in new window) is a cloud-based software-as-a-service (SaaS) that optimises slide evaluation and significantly reduces the time and cost of histological analyses while enhancing diagnostic accuracy. EU funding of the PATH-TOX project supported significant enhancements to the prototype and adoption by world-renowned pharmaceutical companies and top contract research organisations (CROs).
Cloud-based and AI-supported histology collaboration environment
Many organisations now rely on cloud-based document-sharing platforms. Pharmaceutical companies and CROs involved in multisite trials are no different. However, until now, the toxicologic pathology workflow was not digitalised. PATH-TOX coordinator Erio Barale-Thomas of Janssen Pharmaceutica(opens in new window) explains: “The SaaS Patholytix Preclinical(opens in new window) is cloud-based for uploading, analysing, annotating and exporting study content. This comprises the scanned histological slides and all the study information necessary to track and interpret slides such as animal identifier, dose and organ.” Patholytix Preclinical allows pathologists to assess the study content as a montage of all slides for insight at the group level and as individual slides for a diagnosis as accurate as obtained from review under the microscope. The secure and collaborative multisite, multi-organisation content sharing also facilitates remote peer review, eliminating slide transport and pathologist travel time and cost. The process is augmented with predictions generated by AI classifiers. “Partner Deciphex’s deep learning techniques are able to highlight any deviations from the normal tissue without explicitly training the AI model on each individual abnormality. This pre-screening enables pathologists to focus their valuable time on the scientific aspects of challenging cases and take more informed decisions like ending a study early, saving time and resources,” states Barale-Thomas. A preliminary analysis comparing conventional microscope processing with Patholytix Preclinical demonstrated an increase in processing speed of up to 42 % on certain tasks. The Food and Drug Administration recommends the assessment of more than 40 different tissues to determine subchronic toxicity. Translating the Patholytix Preclinical time savings to the salary equivalent of expert toxicologic pathologists could significantly impact the overall cost of nonclinical studies.
PATH-TOX accelerates innovation
The strengthened collaboration during PATH-TOX resulted in deeper understanding and significant improvements. AI processing was performed on brain, heart, kidney, liver and lung. Barale-Thomas adds: “We also found that organs with complex architecture such as the kidneys benefit more from the multi-magnification view provided to the AI analysis than homogenous organs. These results have been published(opens in new window). Further, we discovered that Deciphex’ models can generalise, identifying abnormalities that were not seen during the training process.” Barale-Thomas concludes: “Patholytix Preclinical is the first tool that supports the toxicologic pathologist during the primary evaluation.” This unique triaging of slides should uniquely streamline slide evaluation. Several of the top 10 global pharmaceutical companies and the top 2 CROs have adopted the technology for routine nonclinical pathology applications, and interest from new organisations continues to grow. Toxicologic pathology has entered a new era.
