Brief Review — Accuracy of Artificial Intelligence-Based Photographic Detection of Gingivitis
DeepLabv3+ for Gingivitis Segmentation
Accuracy of Artificial Intelligence-Based Photographic Detection of Gingivitis,
DeepLabv3+ for Gingivitis Segmentation, by The University of Hong Kong, Guangdong University of Technology, The National University of Malaysia, Hong Kong Chu Hai College
2023 Elsevier IDJ, Over 20 Citations (Sik-Ho Tsang @ Medium)Biomedical/Medical Image Segmentation
2022 [UNETR] [Half-UNet] [BUSIS] [RCA-IUNet] [Swin-Unet] [DS-TransUNet] [UNeXt] [AdwU-Net] [TransUNetV2] [Swin-SFTNet] 2023 [DCSAU-Net] [RMMLP] [BTS-ST]
Summary: My Healthcare and Medical Related Paper Readings and Tutorials
==== My Other Paper Readings Are Also Over Here ====
- Gingivitis is one of the most prevalent plaque-initiated dental diseases globally.
- Frontal view intra oral photographs fulfilling selection criteria were collected, with gingivitis condition labeled as healthy, diseased, questionable, or background for each pixels.
- DeepLabv3+ is used for image segmentation for monitoring of the effectiveness of patients’ plaque control.
Outline
- Data Collection
- DeepLabv3+
- Results
1. Data Collection
Consecutive participants were recruited amongst patients attending the Comprehensive Dental Clinic of the University Dental Hospital from 2020 to 2022 according to the selection critera as shown above.
- Frontal view intraoral photographs were taken using a digital single-lens reflex (SLR) camera (EOS 700D, Canon) with a macro lens (EF 100mm f/2.8, Canon) and a ring flash (Marco Ring Lite MR-14EX, Canon).
The sample size used for training the AI system was based on a recent study on using AI to detect periodontitis, which featured around 450 training datasets.
- The gingival conditions of the collected intraoral photographs were labelled by a calibrated assessor, who was a dentist, and based on visual assessment on a computer monitor (P2419H 23.8” W-LED monitor, Dell):
- Healthy: pink, smooth, no bleeding
- Questionable: red, rough, swollen
- Diseased: white/red patches, generalised redness, ulcers, swollen, bleeding
- Background: Remaining pixels.
- One week later, 10% of all photographs were labelled again by the same assessor to measure the intra-assessor reliability in diagnosis of gingival conditions (healthy, diseased, or questionable). The kappa value of 0.92 was measured which shows high reliability of assessor.
Around 450 photographs were randomly designated as training datasets by randomisation table, and the rest of the photographs were designated as validation datasets.
- Photographs of the training datasets were augmented by cropping, rotating, or flipping randomly to enhance the training quality.
2. DeepLabv3+
DeepLabv3+ is used with Xception network as backbone and also with the segmentation head after the backbone, as shown above.
- NVIDIA RTX 3090 is used. 30000 iterations are used. Batch size is 4.
3. Results
With the above results, sensitivity and specificity at or above 0.90 are obtained.
- The above shows some of the examples.
