Field validation of deep learning based Point-of-Care device for early detection of oral malignant and potentially malignant disorders

Abstract

Early detection of oral cancer in low-resource settings necessitates a Point-of-Care screening tool that empowers Frontline-Health-Workers (FHW). This study was conducted to validate the accuracy of Convolutional-Neural-Network (CNN) enabled m(mobile)-Health device deployed with FHWs for delineation of suspicious oral lesions (malignant/potentially-malignant disorders). The effectiveness of the device was tested in tertiary-care hospitals and low-resource settings in India. The subjects were screened independently, either by FHWs alone or along with specialists. All the subjects were also remotely evaluated by oral cancer specialist/s. The program screened 5025 subjects (Images: 32,128) with 95% (n = 4728) having telediagnosis. Among the 16% (n = 752) assessed by onsite specialists, 20% (n = 102) underwent biopsy. Simple and complex CNN were integrated into the mobile phone and cloud respectively. The onsite specialist diagnosis showed a high sensitivity (94%), when compared to histology, while telediagnosis showed high accuracy in comparison with onsite specialists (sensitivity: 95%; specificity: 84%). FHWs, however, when compared with telediagnosis, identified suspicious lesions with less sensitivity (60%). Phone integrated, CNN (MobileNet) accurately delineated lesions (n = 1416; sensitivity: 82%) and Cloud-based CNN (VGG19) had higher accuracy (sensitivity: 87%) with tele-diagnosis as reference standard. The results of the study suggest that an automated mHealth-enabled, dual-image system is a useful triaging tool and empowers FHWs for oral cancer screening in low-resource settings.

Status and Determinants of Early Detection of Oral Premalignant and Malignant Lesions in India

“…A multicentric prospective study reported high accuracy for telediagnosis employing an automated mHealth-enabled, dual-image device as a point-of-care triaging tool in comparison with onsite specialists (sensitivity:95%; specificity:84%), which will empower CHWs for defining OPMD. 59 A previous study has reported enhanced compliance with workplace screening by mobile health. 61 Multi-centric study from Tamil Nadu reported that mobile technology was user-friendly, inspires CHW confidence, and decreases paper-based system time, resources, and maintenance.…”

“…Recent indigenous low-cost innovations such as point-of-care diagnosis layered with artificial intelligence and machine learning are emerging adjunct techniques to improve OPMD detection and biopsy. 59 The Indian Council of Medical Research provides early detection training for dentist; more platforms are needed for standardized training.…”

“…In the Kerala and Karnataka studies, screening by trained CHWs reported moderate sensitivity for detecting OPMD and oral cancer (67.4% and 60%, respectively). 5 , 59 Misclassification in OVE is common due to heterogeneity of oral mucosal lesions. Presence of oral epithelial dysplasia is the most important risk factor for malignant transformation and can be classified as “high-risk” (moderate-18%; severe-39%), and “low-risk” (mild-6%).…”

Machine learning for detection and classification of oral potentially malignant disorders: A conceptual review

“…Previous studies have developed systems using simple photographs, 27–29 smartphone use, 23–25,30,31 and web application 32,34 . For instance, Warin and collaborators reported a pipeline composed of segmentation and classification CNNs capable to detect and classify OPMDs in oral photographs.…”

“…Moreover, the research group also evidenced that phone‐integrated CNN and Cloud‐based CNN had a higher accuracy (sensitivity: 87%) when telediagnosis was employed as standard. Finally, they concluded it is useful to triage and direct oral cancer screening 34 …”