The catalyst effect of Covid-19 on healthcare systems worldwide has further emphasised the need to expedite the application of digital strategies in hospitals, particularly artificial intelligence (AI) pathologic response (PathR) assessment in precision medicine and AI models for patient stratification in routine care. PathR, specifically major pathologic response (MPR), is manually determined by pathologists from tumour resection slides to elucidate the efficacious clinical endpoint (≤10% viable residual tumour) of cancer patients treated with therapy. This is a time-consuming and occasionally erroneous process that can have detrimental impacts on patients’ health outcomes due to treatment flow depending on pathological assessment.

To offset these limitations and gain more accurate metrics than can be achieved by manual evaluation, numerous machine learning (ML) and deep learning programmes are being explored as scalable, precise and reproducible approaches for patient stratification and predicting long-term outcomes. At this year’s American Society of Clinical Oncology (ASCO) conference, Owkin, a startup pioneering AI technologies for medical research and clinical development, presented the clinical utility of the first AI-based tool in predicting the genomic subtypes of pancreatic cancer from histology slides. Owkin’s ML model opens the potential to improve routine care and clinical trial efficiency through early patient molecular stratification, thereby identifying patients who respond better to specific investigative therapies.

Another leading AI vendor providing advanced platforms for clinical development at ASCO 2021 is PathAI, whose collaboration with Roche’s Genentech LCMC3 Phase II study highlighted the potential of ML in supporting pathologists to assess PathR. PathAI’s digital model analysed the pathological specimens from non-small cell lung cancer patients who had resections following neoadjuvant use of Roche’s Tecentriq (atezolizumab) and revealed a strong agreement between its ML-model assessment of MPR and its manual comparator arm (area under the receiver operating characteristic = 0.975). In addition, digitally assessed MPR was significantly associated with better disease-free survival (DFS) and overall survival (OS), while its manual comparator arm showed a non-significant trend towards better DFS and OS.

Although further data maturation of the LCMC3 trial is still required to make concrete conclusions, PathAI’s multi-faceted platform sets itself apart as a patient enrolment and candidate companion diagnostics tool generating reproducible pathology scores into clinical trial workflows. In May, PathAI raised $165m in its Series C bid to expand its clinical research capabilities. Contributing investors included Bristol Myers Squibb, Labcorp and Merck’s Global Health Innovation Fund.

Though not all were at ASCO 2021, other worthy mentions are Deep Lens, AiCure, GNS and Unlearn.AI, whose platforms also focus on identifying appropriate patients, predicting patient outcomes and streamlining the operational process of clinical trials.

While there are ample opportunities for business traction in the digital pathology market, a major limitation envisioned would be the lack of ultrafast scanners and image management systems in several hospitals worldwide. This, coupled with the willingness of pathologists to revise and adopt digitalised image analysis, could impede the market traction of digital pathology. Factors like Covid-19 and the continual rise of cancer worldwide have, however, further emphasised the need for imminent adoption of AI to improve patient stratification in routine care and clinical trials, predict outcomes and decrease histopathology turnaround time.

GlobalData forecasts from the AI in Healthcare Thematic report published in April suggest that the market for AI platforms for the entire healthcare industry will reach $4.3bn by 2024, up from $1.5bn in 2019. Compound annual growth rates (CAGRs) between 2019 and 2024 will be 24.6% in healthcare, 18.5% in pharma and 20.6% in medical devices.