Artificial intelligence tool for cassava viral diseases diagnosis using participatory surveillance in Burkina Faso

Seydou Sawadogo^1,2Fidele Tiendrebeogo^1,2,3*Ezechiel B Tibiri^1,2Pakyendou E Name^1,2Florencia Djigma²Lassina Traoré²Justin S Pita³Angela O Eni³

• ¹Institut de l'environnement et de la Recherche Agricole (INERA), Ouagadougou, Burkina Faso
• ²Joseph KI-ZERBO University, Ouagadougou, Kadiogo, Burkina Faso
• ³Central and West African Virus Epidemiology (WAVE), Bingerville, Côte d'Ivoire

In the area of plant health, there has been little work using participatory approaches to control emerging infectious diseases such as cassava mosaic disease (CMD) and cassava brown streak disease (CBSD). These diseases cause significant yield losses in Sub-Saharan Africa. The current study provided low cost and early detection method of cassava viral diseases surveillance, based on participatory approaches using an AI tool (Plantvillage nuru app). The study involved farmers, agricultural extension agents (AEA), and experts cassava diseases diagnosis experts. Farmers were made aware of CMD and CBSD damage through a national campaign, while AEA received training to identify CMD, CBSD, and cassava green mite (CGM) symptoms using an AI-based diagnostic tool. Sixty trained AEA, equipped with smartphones running the AI tool, conducted fields surveillance either through visual inspection or with AI tool. The participation rate of the AEA and the diagnostic accuracy of the AI tool and visual assessments were evaluated and compared to experts perception validated by molecular analysis. Workshops and smartphones allocation enhanced AEA participation rate to 60%, and increased surveyed fields number to 132. CMD detection revealed no significant difference between users of AI tool (P-value = 0.709) and visual inspection (P-value = 0.997). The mean scores of CMD detection were 29.83 ± 12.99% for AI tool, 37.12 ± 12.78% for experts, and 36.10 ± 12.74% for molecular analysis among AI tool users. With visual inspection users, the mean scores detection were 46.07 ± 13.00% for AEA and experts perception, and 43.87 ± 12.07% for molecular analysis. The AI tool misdiagnosed 5% of CMD as CBSD, but molecular analysis confirmed it as CMD. The CMD infected fields was 31.06%, with a predominantly African Cassava Mosaic Virus (93.33%) detected. The results demonstrated that participatory approaches could be effective in the plant pathogens early management.