Integrated Molecular and AI-Based Diagnostics for Banana Diseases: Development, Optimization, and Field Deployment of LAMP and Computer Vision Technologies

Abstract

Banana and plantain (Musa spp.) production in Sub-Saharan Africa is severely constrained by multiple diseases, with Banana bunchy top virus (BBTV) representing the most devastating viral threat. Inadequate diagnostic infrastructure limits effective management, particularly for asymptomatic infections disseminated through informal planting material exchange. This study presents an integrated diagnostic framework combining Loop-Mediated Isothermal Amplification (LAMP) molecular diagnostics with deep learning-based computer vision for rapid, scalable disease detection under field conditions. A LAMP assay targeting the BBTV DNA-S coat protein gene was developed using conserved sequences from diverse African isolates and validated with a simplified alkaline extraction protocol eliminating conventional DNA purification. The assay achieved 100% specificity and concordant detection with PCR and qPCR, reducing diagnostic time from 4 to 6 hours to 60 minutes. In-house recombinant Bst LF polymerase production demonstrated comparable enzymatic performance to commercial alternatives, with projected per-reaction cost reductions of 70 to 80%. Concurrently, an SSDLite MobileNetV2 object detection model was developed through 19 iterative training cycles on 19,914 field-collected images spanning 22 disease and physiological stress classes. The final model achieved recall rates of 92.5% for BBTV, 91.0% for Banana Xanthomonas Wilt, and 98.1% for healthy leaf classification, deployed via the PlantVillage mobile application for real-time offline diagnostics. A QR code-based metadata system integrates phenotypic AI assessments with molecular confirmation for comprehensive surveillance. This complementary framework addresses broad-scale phenotypic screening and molecular confirmation of pre-symptomatic infections, providing accessible tools to safeguard food security across Sub-Saharan Africa.

Plant phenotyping relevance

植物病害の表現型を画像から推定するコンピュータビジョン手法の開発・評価・アプリ展開が中心であり、LAMPによる分子診断も補完的に統合されている。

Code and data availability

The supplied blocks describe a field-collected banana disease image dataset (19,914 images), an SSDLite MobileNetV2 model, and LAMP assay development, but contain no public deposit, availability statement, or authors' URL for the dataset, images, annotations, trained model, or analysis code. The only URL mentioned (NEB

No evidence-backed public reproduction asset is currently recorded.