Retinex-Based Hybrid CNN–LSTM Architecture for Periapical Lesion Detection in CBCT and Panoramic Radiographs
DOI:
https://doi.org/10.58466/vc18dh25Keywords:
Deep Learning, CNN-LSTM, Retinex, Periapical Lesions, Cone-Beam Computed Tomography (CBCT), Panoramic RadiographyAbstract
Periapical lesion detection plays a crucial role in endodontic diagnosis; however, manual interpretation of Cone-Beam Computed Tomography (CBCT) and panoramic radiographs remains time-consuming, highly dependent on the clinician's expertise, and susceptible to diagnostic variability. This study proposes a hybrid deep learning framework that integrates a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture, combined with Retinex-based image enhancement, for the automatic detection and classification of periapical lesions. Retinex enhancement is employed as a preprocessing step to normalize illumination and improve lesion contrast. The hybrid CNN-LSTM model captures both spatial and contextual dependencies through sequential patch-based processing of panoramic and CBCT images. Using a dataset of 1,500 annotated images collected from clinical radiographic datasets and publicly available sources, the proposed model achieved an accuracy of 97.8%, precision of 96.4%, recall of 95.9%, and an F1-score of 0.96, significantly outperforming conventional CNN and U-Net models. These findings demonstrate that the integration of image enhancement and hybrid deep learning improves sensitivity to small lesions while reducing false-negative detections, offering a clinically viable AI-assisted approach for endodontic diagnosis.
References
[1] P. Patel et al., “Cone-beam computed tomography in endodontics,” Int. Endod. J., vol. 52, pp. 1360-1373, 2019.
[2] S. Estrela et al., “Characterization of periapical lesions by cone-beam CT,” Oral Surg. Oral Med. Oral Pathol. Oral Radiol. Endod., vol. 106, pp. 909-914, 2008.
[3] H. Setzer et al., “Comparison of periapical diagnosis using CBCT and conventional radiography,” J. Endod., vol. 44, pp. 1091-1098, 2018.
[4] J. Ver Berne et al., “A deep learning approach for radiological detection and classification of radicular cysts and periapical granulomas,” J. Dent., vol. 135, 2023.
[5] K. Chau et al., “A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM,” J. Dent., vol. 153, 2025.
[6] M. Allihaibi et al., “Diagnostic accuracy of an AI-based platform in detecting periapical radiolucencies on CBCT scans,” J. Dent., vol. 160, 2025.
[7] A. Lee et al., “Artificial intelligence in dentistry: Current applications and future perspectives,” Dent. Mater. J., vol. 42, no. 1, pp. 13-25, 2023.
[8] S. Yang et al., “Development of a visually explainable deep learning model for classification of C-shaped canals,” J. Endod., vol. 48, no. 7, pp. 914-921, 2022.
[9] K. Wang et al., “Deep learning-based efficient diagnosis of periapical diseases with dental X-rays,” Image Vis. Comput., vol. 147, 2024.
[10] J. Ver Berne et al., “Automated classification of panoramic radiographs with inflammatory periapical lesions using a CNN-LSTM architecture,” J. Dent., vol. 156, 2025.
[11] A. Khan et al., “Hybrid CNN-RNN architectures for medical image classification: A systematic review,” Comput. Biol. Med., vol. 153, 2023.
[12] D. Jobson et al., “Properties and performance of a center/surround Retinex,” IEEE Trans. Image Process., vol. 6, no. 3, pp. 451-462, 1997.
[13] V. Latke and V. Narawade, “Detection of dental periapical lesions using Retinex based image enhancement and lightweight deep learning model,” Image Vis. Comput., vol. 146, 2024.
[14] A. Choudhury et al., “Improved medical image contrast enhancement using multi-scale Retinex with adaptive gamma correction,” Biomed. Signal Process. Control., vol. 81, 2023.
[15] M. Fu et al., “Fully 3D PAL-Net for periapical lesion segmentation,” Dentomaxillofacial Radiol., vol. 51, no. 2, 2022.
[16] K. Wang et al., “Mask R-CNN for automated detection and classification of periapical diseases,” Image Vis. Comput., vol. 147, 2024.
[17] R. Rahman et al., “A survey on deep learning in medical image analysis,” Med. Image Anal., vol. 69, 2021.
[18] J. Fawcett, “An introduction to ROC analysis,” Pattern Recognit. Lett., vol. 27, no. 8, pp. 861-874, 2006.
[19] L. S. Tomita et al., “Interobserver agreement in periapical lesion detection using CBCT,” Clin. Oral Invest., vol. 28, no. 4, pp. 1571-1578, 2024.
[20] K. Chau et al., “A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM,” J. Dent., vol. 153, 2025.
[21] A. Raj et al., “Data augmentation strategies for improving dental radiograph analysis,” Comput. Biol. Med., vol. 161, 2023.
[22] Z. Li et al., “Contrast enhancement using multiscale Retinex for medical image segmentation,” Pattern Recognit. Lett., vol. 155, 2022.
[23] Y. Zhang et al., “Illumination normalization with Retinex for dental panoramic image analysis,” Med. Phys., vol. 49, no. 7, pp. 4776-4788, 2022.
[24] P. Shukla et al., “Sequential learning in hybrid CNN-LSTM models for medical image diagnosis,” IEEE Access, vol. 11, pp. 25839-25852, 2023.
[25] E. Park et al., “Temporal feature modeling in dental radiographs using CNN-RNN networks,” Dentomaxillofacial Radiol., vol. 52, no. 4, 2023.
[26] A. El-Nouby et al., “Noise-resilient deep learning for clinical image analysis,” Pattern Recognit., vol. 139, 2023.
[27] S. Seegerer et al., “Evaluating interpretability and robustness of CNNs in radiology,” Med. Image Anal., vol. 88, 2024
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Safar Dwi Kurniawan, Tri Haryo Nugroho, David Bani Adam
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
