BackgroundPernicious anemia (PA) is a severe clinical consequence of autoimmune gastritis. It results from immune-mediated damage to gastric parietal cells in the oxyntic mucosa.
This process leads to intrinsic factor deficiency and subsequent vitamin B12 malabsorption. This complex autoimmune response, combined with non-specific clinical manifestations, generates substantial diagnostic uncertainty.
This uncertainty frequently results in misdiagnosis or delayed diagnosis and, consequently, multiple adverse outcomes, including irreversible neurological complications.MethodsTo address the intrinsic diagnostic uncertainty of PA—arising from heterogeneous, graded, and often discordant clinical, histological, immunological, and biochemical information—we developed ISPAD (Intelligent System for Pernicious Anemia Diagnosis), an explainable AI–based probabilistic framework to integrate this information, producing a probability estimate of pernicious anemia that reflects clinical reasoning rather than rigid diagnostic thresholds.ResultsISPAD was examined using a series of published diagnostically challenging cases reflecting real-world complexity. These cases included antibody assay interference, hemolysis-masked macrocytosis, seronegative presentations, and cancer-associated atrophy.
Across cases, the system generated a continuous and adaptable probabilistic assessment of pernicious anemia.
Pernicious anemia (PA) arises from autoimmune-mediated injury to gastric oxyntic mucosa leading to intrinsic factor deficiency and impaired vitamin B12 absorption.
The disorder commonly presents with nonspecific features and a mosaic of clinical, histological, immunological, and biochemical findings, which together create diagnostic ambiguity and frequent mis- or delayed diagnoses.
The authors sought to reduce diagnostic uncertainty by creating a computational framework—ISPAD (Intelligent System for Pernicious Anemia Diagnosis)—that models expert clinical reasoning.
Rather than applying fixed cutoffs, the system produces a continuous probability of PA by integrating multiple heterogeneous data types within an explainable, probabilistic architecture.
ISPAD is described as an explainable AI–based probabilistic model designed to assimilate graded and potentially conflicting inputs (clinical presentation, histology, serology, biochemical markers).
The framework emphasizes transparency of inference so that probability estimates reflect context-dependent weighting akin to clinician judgement.
Specific algorithmic details, training procedures, validation cohorts, and performance metrics were not reported in the source.
The system was evaluated on a set of published, diagnostically challenging case vignettes intended to mirror real-world complexity—including antibody assay interference, hemolysis-concealed macrocytosis, seronegative disease, and gastric atrophy related to malignancy.
ISPAD yielded continuous, adaptable probability outputs across these scenarios, demonstrating capacity to formalize multifactorial reasoning rather than binary classification.
As a proof-of-concept, ISPAD exemplifies how explainable AI can structure complex diagnostic assessment in autoimmune-related PA by integrating discordant evidence into a transparent probabilistic estimate.