An Evaluation Method of Dental Treatment Quality Combined with Deep Learning and Multi-index Decomposition

Figures & data

Figure 1. Dental treatment quality evaluation process.

Figure 2. CenterNet network structure.

Figure 3. U-net network structure.

Figure 4. Root canal data set after data augmentation.

Figure 5. Result of root canal image segmentation.

Figure 6. Comparison of preoperative and postoperative images of root canal therapy and filling teeth.

Figure 7. Scoring index graph of filling teeth.

Table 1. The server environment used in the experiment.

Type	Configuration
Processor	Intel(R) Xeon(R) CPU @ 2.20 GHz
Memory	256 GB
Graphics card	Tesla T4
Display memory	16GB
GPU library	CUDA11.2 and CUDNN7.6
Deep learning framework	Tensorflow 1.13.2

Table 2. Comparison of three main feature extraction networks.

Backbone network	mAP/%	FPS/(frame/s)
ResNet50	96.26	21.7
ResNet101	94.70	14.4
Hourglass	95.18	4.7

Figure 8. The comparison of segmentation effect between marked watershed and U-net.

Figure 9. Software scoring interface display.

Table 3. Comparison of root canal treatment quality evaluation results.

	Algorithm	Doctor 1	Doctor 2	Doctor 3
Patients	Score/Grade	Score/Grade	Score/Grade	Score/Grade
Patient 1	93/A	90/A	81/B	95/A
Patient 2	84/B	76/B	90/A	81/B
Patient 3	92/A	79/B	87/B	88/B
Patient 4	80/B	87/B	84/B	32/C
Patient 5	86/B	89/B	94/A	91/A
Patient 6	35/C	84/B	87/B	75/B
Patient 7	85/B	75/B	80/B	82/B
Patient 8	82/B	87/B	73/B	28/C
Patient 9	28/C	24/C	71/B	26/C
Patient 10	88/B	72/B	73/B	83/B

Table 4. Comparison of dental filling quality evaluation results.

	Algorithm	Doctor 1	Doctor 2	Doctor 3
Patients	Score/Grade	Score/Grade	Score/Grade	Score/Grade
Patient 1	83.3/B	73.3/C	81.3/B	76.3/C
Patient 2	84.0/B	78.0/C	90.0/A	77.0/C
Patient 3	82.0/B	79.0/C	87.7/B	76.0/C
Patient 4	80.7/B	87.7/B	84.7/B	74.0/C
Patient 5	86.0/B	89.0/B	94.0/A	81.0/B
Patient 6	85.3/B	84.3/B	87.3/B	86.7/B
Patient 7	78.0/C	74.0/C	85.0/B	75.0/C
Patient 8	88.0/B	83.0/B	92.0/A	88.7/B
Patient 9	83.0/B	65.0/C	78.0/C	85.0/B
Patient 10	92.3/A	90.0/A	95.0/A	94.0/A

Table 5. Average time-consuming comparison of each picture.

Our algorithm	Doctor 1	Doctor 2	Doctor 3	The average time consumed by doctors
0.147s	20.1s	16.3s	27.0s	21.1s