Aiforia's tech advances to expedite path to profitability

Aiforia will publish H2’25 results on March 6. We view FY26 as an inflection year, with stronger revenue growth and operational efficiencies accelerating the path to profitability.

Aiforia recently launched its proprietary Foundation Engine, a patent-pending technology based on the Vision Transformer (ViT) architecture. The ViT architecture is seen to adapt well to the heterogeneous, challenging features of whole slide images. It is a relatively novel architecture, challenging Convolutional Neural Networks (CNN) long-standing dominance in medical imaging. Through global self-attention, ViTs capture the broader context of an image more effectively than local convolutions. Takahashi et al. (2024) reviewed 36 medical imaging studies (X-rays, CT scans, MRIs, histopathology slides), comparing ViTs/Transformers to CNNs, finding ViTs performed generally better or comparably in many classification tasks (e.g. skin cancer, ViT 94.3% accuracy).

Aiforia's Foundation Engine exemplifies this ViT advantage, but it also highlights the fundamental trade-offs between Foundation Models (FM) and traditional CNNs in pathology workflows. Pre-trained on vast unlabeled datasets, FMs deploy and scale rapidly while generalising well across a vast spectrum of cases, but trade cell-level granularity for broader slide context at the cost of slower, compute-heavy inference. CNNs on the other hand, deliver precise cell-level detail more efficiently, yet struggle in novel cases outside their training data, requiring vast annotated datasets, which pathology desperately lacks. Therefore, training AI models with CNNs can prove difficult, costly and time-consuming.

Aiforia sees it as essential to bridge both worlds for clinical use, combining Foundation Models' broad generalisation with task-specific fine-tuning for validated, precise performance. Foundation Engine-trained models should generalise well across different cancers and scanners, while providing cell-level precision with leaner, faster inference than pure FMs. By leveraging pre-trained FMs, the Foundation Engine gains a "headstart" that requires fewer annotated slides for training, which accelerates development and slashes costs. We expect Foundation Engine models to exhibit improved accuracy in edge cases (vs CNN), where even experienced pathologists struggle to reach consensus. Overall, Foundation Engines will drive faster deployment, superior adaptability, lower cost, and enhanced clinical scalability. Noteworthy, the latest CE-IVD marked gastric cancer solution was built using the Foundation Engine.

We remain upbeat on Aiforia’s Foundation Engine and its clinical and commercial scalability. Aiforia’s clinical footprint now covers breast, prostate, and lung cancers, strengthened by its new gastric cancer model. Supply–demand dynamics are supportive as hospitals face rising pressure to digitise pathology and adopt AI. In Lombardia, ~400 pathologists in 2025 are expected to fall by ~100 (due to retirements) by 2027 with only nine replacements, underscoring urgent automation needs. FY26 should bring additional tenders, including HUS, Finland’s largest hospital group (potentially in Q2). We reiterate our BUY rating with a € 3.80 PT based on our DCF model.