Next-generation Diagnostics for Silkworm Diseases: A Critical Narrative Review of Molecular Markers, Biosensors, Imaging and Artificial Intelligence

Abstract

Diseases of the domesticated silkworm, Bombyx mori, remain a principal cause of cocoon crop loss, and the inspection systems that protect silkworm egg production still depend heavily on light microscopy and visible symptoms. During the past decade, nucleic-acid amplification, clustered regularly interspaced short palindromic repeats-based detection, electrochemical and lateral-flow biosensing, vibrational spectroscopy, quantitative imaging and machine learning have each been proposed as faster or more sensitive alternatives. Existing reviews have generally treated these technologies separately, and none has judged them against an explicit framework of intended use, reference-standard quality and field validation. This critical narrative review integrates peer-reviewed evidence on diagnostics for microsporidian pébrine, nucleopolyhedrovirus grasserie, the viral and bacterial flacherie complex and fungal muscardine, drawing on international bibliographic databases, regional Japanese and Korean scholarly platforms and citation tracking. The evidence indicates that quantitative polymerase chain reaction is the only modality evaluated at the scale of routine egg-production inspection, where it agreed closely with female moth microscopy on positive samples while detecting additional infections that microscopy missed. Isothermal and clustered regularly interspaced short palindromic repeats-based assays have reached high analytical sensitivity and increasingly portable formats, but most evaluations rely on plasmids, purified genomic DNA or experimentally infected larvae rather than field lots. Electrochemical biosensors report very low detection limits in buffered or spiked samples, yet independent replication and comparison with reference methods are scarce. Spectroscopic and image-based classifiers frequently report accuracies above ninety per cent, although small single-site datasets, visually assigned labels and absent external validation limit confidence in their transferability. Pathogen diversity, latent and mixed infections, and the imperfect nature of microscopy as a reference standard affect every modality. The available evidence supports qualified adoption of quantitative and isothermal nucleic-acid assays for seed inspection, whereas biosensors and artificial intelligence remain at the proof-of-concept stage. Progress now depends on fitness-for-purpose validation, multi-site sample panels with molecularly confirmed labels, lot-level sampling designs and tiered diagnostic architectures that link non-invasive screening to confirmatory testing.