The architecture translates foundation model outputs into pathologist-validated tissue structures, with detection algorithms showing up to 99.85% negative predictive value in routine casework.


Deciphex announced the public launch of CipherX, an artificial intelligence engine designed to provide auditable, model-invariant infrastructure for histopathology workflows across clinical diagnostics and research. The platform has been in active production since January 2026, serving as the core computational engine behind the company’s Diagnexia clinical diagnostic service and its Patholytix pharmaceutical research business.

According a press release from Deciphex, the engine addresses key limitations associated with raw pathology foundation models, which convert whole-slide image tiles into high-dimensional numerical embeddings. While industry development has focused on expanding dataset sizes over the past three years, performance at the tile level has largely converged without resolving core diagnostic hurdles, such as tracing predictions back to visible morphological features on a glass slide, the release states. Furthermore, regulatory bodies increasingly demand auditable systems that do not require downstream tools to be redeveloped when an underlying foundation model is updated or replaced, Deciphex notes.

“There are strong pathology foundation models available today, and more coming from well-resourced teams,” says Dr Donal O’Shea, of Deciphex, in a release. “What the field has been missing is a stable layer above the models that speaks in pathology terms, holds up as models change underneath, and lets a pathologist see how a conclusion was assembled. CipherX is that layer, and it has been running our clinical and research work for months.”

Two-Component Architecture and Semantic Layer

CipherX is engineered around two primary components: a foundation model layer and a proprietary semantic layer, according to the release. The foundation model layer incorporates multiple neural network encoders, including Deciphex’s proprietary DCX-3 fusion model, and dynamically selects an encoder based on specific diagnostic tasks and data-use rights, the company stated. On the THUNDER and EVA public benchmark suites, the DCX-3 model ranks alongside prominent pathology foundation models, including Virchow2, UNI2-H, H-Optimus-1, RudolfV2, and Midnight, Deciphex reported.

Positioned directly above the encoders, the semantic layer decomposes mathematical embeddings into discrete units called “glyphs,” which represent recurring structural tissue elements confirmed and named by subspecialist pathologists, according to Deciphex. These glyphs assemble into recognizable histological signatures, allowing the system to construct higher-order spatial arrangements—such as tumor-infiltrating lymphocytes or tertiary lymphoid structures with germinal-center architecture—without initiating new model training cycles, the company noted. Because each output links directly to named structures that a reviewing pathologist can inspect and reject, substituting the underlying foundation model does not alter the validated clinical vocabulary established with laboratory customers, according to the release.

Clinical Laboratory and Research Applications

Within routine clinical operations, Deciphex deploys CipherX through three Diagnexia Assist tools designed to operate outside the primary diagnostic pathway, focusing on pre-analytical image quality assessment, complexity-based case triage, and post-authorization quality review, the company states.

In diagnostic casework, detection algorithms built on CipherX signatures have achieved a 99.85% negative predictive value for adenocarcinoma and a 98.76% negative predictive value for melanoma in routine casework, according to company performance data.

For life sciences and translational pathology, the engine supports digital tissue biomarker development and digital companion diagnostics via Patholytix Research Services, Deciphex reported. Any biomarkers generated using the platform maintain the same underlying framework, ensuring that analytical findings can be reviewed and audited against identifiable tissue morphology, the company notes.

Deciphex states that CipherX is not being marketed as a standalone licensed software product; instead, it is delivered directly within Diagnexia clinical workflows, embedded in the Patholytix platform, or deployed through custom development programs for study sponsors. The company emphasized that the system does not substitute for professional medical judgement, with all active production tools functioning as adjuncts to assist pathologist decision-making.

ID 25712658 © Xunbin Pan | Dreamstime.com