Sanyam Jain
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Sanyam Jain has a total of 38 co-authors. Here are the top co-authors by number of shared publications:
- Ruben Pauwels
Department of Dentistry and Oral Health Aarhus University Aarhus Denmark
8 shared publications - Khuram Naveed
Department of Dentistry and Oral Health Aarhus University Aarhus Denmark
3 shared publications - Ahmad Badruddin Ghazali
Oral Radiology Unit, International Islamic University Malaysia, Pahang, Malaysia
2 shared publications - Akhilanand Chaurasia
Department of Oral Medicine and Radiology, King George’s Medical University, Lucknow, India
2 shared publications - Alexandros Iosifidis
Tampere University, Tampere, Finland
2 shared publications - Basel Khalil
Department of Dental Medicine, Karolinska Institute, Stockholm, Sweden; ITU/WHO/WIPO Global Initiative on Artificial Intelligence for Health, Topic Group Oral Health, Geneva, Switzerland
2 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (10) , 496
Self-Replicating Neural Cellular Automata: Quantifying Emergent Phenotypic and Genotypic Diversity in an Open-Ended Substrate
We study an in silico substrate in which every pixel of a two-channel cellular-automata grid carries a tiny neural network (an agent) that senses its Moore neighbourhood. A cell persists only by self-replication: a living neighbour is cloned, and its weights are mutated by a uniform perturbation, so that phenotype (cell state) is driven entirely by genotype (network weights). From a handful of seeded founders, the system grows into a spatially organized ecosystem of coexisting, competing and dominating species. Our main contribution is a battery of coarse-grained diversity metrics that make such growth measurable at two scales: four phenotypic tools based on cellular-type frequency, entropy and cell variance, and two genotypic tools that colour each agent by a hash of its full weight vector versus a sparse random-weight probe. Across a five-fold sweep of 1680 small runs and 24 long (1000-generation, 200 × 200) runs, the substrate is persistent and self-maintaining in 20 of the 24 long configurations and exposes a clear phenotype–genotype diversity trade-off: raising phenotypic diversity collapses genotypic diversity and vice versa. Full-genome hash colouring further reveals lineage structure that a random-weight probe systematically misses.
- 9 October 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (10) , 496
Open AccessSelf-Replicating Neural Cellular Automata: Quantifying Emergent Phenotypic and Genotypic Diversity in an Open-Ended Substrate
We study an in silico substrate in which every pixel of a two-channel cellular-automata grid carries a tiny neural network (an agent) that senses its Moore neighbourhood. A cell persists only by self-replication: a living neighbour is cloned, and its weights are mutated by a uniform perturbation, so that phenotype (cell state) is driven entirely by genotype (network weights). From a handful of seeded founders, the system grows into a spatially organized ecosystem of coexisting, competing and dominating species. Our main contribution is a battery of coarse-grained diversity metrics that make such growth measurable at two scales: four phenotypic tools based on cellular-type frequency, entropy and cell variance, and two genotypic tools that colour each agent by a hash of its full weight vector versus a sparse random-weight probe. Across a five-fold sweep of 1680 small runs and 24 long (1000-generation, 200 × 200) runs, the substrate is persistent and self-maintaining in 20 of the 24 long configurations and exposes a clear phenotype–genotype diversity trade-off: raising phenotypic diversity collapses genotypic diversity and vice versa. Full-genome hash colouring further reveals lineage structure that a random-weight probe systematically misses.
- Open Access
- 17 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (12) , 6154
Deep Learning-Based Assessment of the Relation Between the Third Molar and Mandibular Canal on Panoramic Radiographs Using Local, Centralized, and Federated Learning in a Simulated Multi-Center Setting
Impaction of the mandibular third molar in proximity to the mandibular canal increases the risk of inferior alveolar nerve injury. Panoramic radiography is routinely used to assess this relationship. Automated classification of molar–canal overlap could support clinical triage and reduce unnecessary CBCT referrals, while Federated Learning (FL) enables multi-center collaboration without sharing patient data. We compared Local Learning (LL), FL, and Centralized Learning (CL) for binary overlap/no-overlap classification on cropped panoramic radiographs partitioned across eight independent labelers in a simulated heterogeneous multi-center setting. A pretrained ResNet-34 was trained under each paradigm and evaluated using per-client metrics with locally optimized thresholds and pooled test performance with a global threshold. Performance was assessed using area under the receiver operating characteristic curve (AUC) and threshold-based metrics, alongside training dynamics, Grad-CAM visualizations, and server-side aggregate monitoring signals. On the test set, CL achieved the highest performance (AUC 0.831; accuracy ≈ 0.782), FL showed intermediate performance (AUC 0.757; accuracy ≈ 0.703), and LL generalized poorly across clients (AUC range ≈ 0.619–0.734; mean ≈ 0.672). Training curves suggested overfitting, particularly in LL models, and Grad-CAM indicated more anatomically focused attention in CL and FL. Overall, centralized training provided the strongest performance, while FL offers a privacy-preserving alternative that outperforms LL.
- 17 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (12) , 6154
Open AccessDeep Learning-Based Assessment of the Relation Between the Third Molar and Mandibular Canal on Panoramic Radiographs Using Local, Centralized, and Federated Learning in a Simulated Multi-Center Setting
Impaction of the mandibular third molar in proximity to the mandibular canal increases the risk of inferior alveolar nerve injury. Panoramic radiography is routinely used to assess this relationship. Automated classification of molar–canal overlap could support clinical triage and reduce unnecessary CBCT referrals, while Federated Learning (FL) enables multi-center collaboration without sharing patient data. We compared Local Learning (LL), FL, and Centralized Learning (CL) for binary overlap/no-overlap classification on cropped panoramic radiographs partitioned across eight independent labelers in a simulated heterogeneous multi-center setting. A pretrained ResNet-34 was trained under each paradigm and evaluated using per-client metrics with locally optimized thresholds and pooled test performance with a global threshold. Performance was assessed using area under the receiver operating characteristic curve (AUC) and threshold-based metrics, alongside training dynamics, Grad-CAM visualizations, and server-side aggregate monitoring signals. On the test set, CL achieved the highest performance (AUC 0.831; accuracy ≈ 0.782), FL showed intermediate performance (AUC 0.757; accuracy ≈ 0.703), and LL generalized poorly across clients (AUC range ≈ 0.619–0.734; mean ≈ 0.672). Training curves suggested overfitting, particularly in LL models, and Grad-CAM indicated more anatomically focused attention in CL and FL. Overall, centralized training provided the strongest performance, while FL offers a privacy-preserving alternative that outperforms LL.
- Open Access
- 20 May 2026
- Published by MDPI AG in Preprints.org
Synthesizing Mandibular Third Molars with Denoising Diffusion Probabilistic Models and Generative Adversarial Networks
In oral surgery, the classification of the proximity of the mandibular third molar to the mandibular canal, typically performed on panoramic radiographs, is essential for surgical planning. While artificial intelligence (AI) tools have been explored for this task, their performance is limited due to data scarcity and class imbalance. In this work, we study the potential of synthetic data generation for this task using Denoising Diffusion Probabilistic Models (DDPMs) and Generative Adversarial Networks (GANs), both unconditional and conditioned to the tooth-canal relationship. We used public datasets to create and label a training dataset of 5416 images. The results show the lowest Fréchet inception distance (FID) / second-highest Inception Score (IS) for the unconditional GAN (32.48 / 2.14). The unconditional DDPM showed an FID of 34.28 and IS of 1.95. Conditional models showed similar IS but a worse overall FID of 68.19 and 219.11 for DDPM and GAN, respectively. In a paired observer study between the two unconditional models, clinical observers found the DDPM image to be more realistic in 69.6% of cases. Future work should investigate downstream effects of GANs and DDPMs used in data augmentation for the training of an AI classifier.
- 20 May 2026
- Published by MDPI AG in Preprints.org
Open AccessSynthesizing Mandibular Third Molars with Denoising Diffusion Probabilistic Models and Generative Adversarial Networks
In oral surgery, the classification of the proximity of the mandibular third molar to the mandibular canal, typically performed on panoramic radiographs, is essential for surgical planning. While artificial intelligence (AI) tools have been explored for this task, their performance is limited due to data scarcity and class imbalance. In this work, we study the potential of synthetic data generation for this task using Denoising Diffusion Probabilistic Models (DDPMs) and Generative Adversarial Networks (GANs), both unconditional and conditioned to the tooth-canal relationship. We used public datasets to create and label a training dataset of 5416 images. The results show the lowest Fréchet inception distance (FID) / second-highest Inception Score (IS) for the unconditional GAN (32.48 / 2.14). The unconditional DDPM showed an FID of 34.28 and IS of 1.95. Conditional models showed similar IS but a worse overall FID of 68.19 and 219.11 for DDPM and GAN, respectively. In a paired observer study between the two unconditional models, clinical observers found the DDPM image to be more realistic in 69.6% of cases. Future work should investigate downstream effects of GANs and DDPMs used in data augmentation for the training of an AI classifier.
- Open Access
Brain2VLM: Hierarchical Alignment Between Cortical Representations and Vision-Language Latent Spaces
This work introduces Brain2VLM, a framework for analyzing how cortical representations align with latent spaces of pretrained diffusion-based vision-language models for brain-to-image reconstruction. While recent approaches achieve strong performance by mapping functional Magnetic Resonance Imaging (fMRI) signals to model latents, the structure of this mapping remains poorly understood. We hypothesize that brain-to-latent alignment is hierarchical, with early visual cortex exhibiting approximately linear correspondence to structural diffusion latents, and higher-order visual areas requiring nonlinear mappings to align with semantic embedding spaces. To test this, we decode diffusion latents and CLIP embeddings from fMRI signals using both linear ridge regression and a nonlinear residual MLP on the Natural Scenes Dataset. Our results reveal that nonlinear decoding provides only marginal improvements for diffusion latents (≈Δ 0.05 − 0.06 in correlation), but yields substantial gains for semantic embeddings (Δ≈0.47), significantly improving distributional alignment (MMD: 0.042 vs 0.358). However, increased decoder expressivity can introduce shifts in latent distributions, highlighting a trade-off between prediction accuracy and generative compatibility. Despite using a simple reconstruction pipeline, Brain2VLM achieves strong performance (PixCorr 0.33, CLIP 85%), suggesting that improvements in brain-to-latent alignment play an important role in reconstruction quality alongside generative modeling. These findings provide empirical evidence for hierarchical alignment between cortical representations and model latent spaces, positioning the brain-to-latent interface as a primary bottleneck in brain decoding systems. Our code can be found at https://github.com/adarsh-crafts/Brain2VLM
Open AccessBrain2VLM: Hierarchical Alignment Between Cortical Representations and Vision-Language Latent Spaces
This work introduces Brain2VLM, a framework for analyzing how cortical representations align with latent spaces of pretrained diffusion-based vision-language models for brain-to-image reconstruction. While recent approaches achieve strong performance by mapping functional Magnetic Resonance Imaging (fMRI) signals to model latents, the structure of this mapping remains poorly understood. We hypothesize that brain-to-latent alignment is hierarchical, with early visual cortex exhibiting approximately linear correspondence to structural diffusion latents, and higher-order visual areas requiring nonlinear mappings to align with semantic embedding spaces. To test this, we decode diffusion latents and CLIP embeddings from fMRI signals using both linear ridge regression and a nonlinear residual MLP on the Natural Scenes Dataset. Our results reveal that nonlinear decoding provides only marginal improvements for diffusion latents (≈Δ 0.05 − 0.06 in correlation), but yields substantial gains for semantic embeddings (Δ≈0.47), significantly improving distributional alignment (MMD: 0.042 vs 0.358). However, increased decoder expressivity can introduce shifts in latent distributions, highlighting a trade-off between prediction accuracy and generative compatibility. Despite using a simple reconstruction pipeline, Brain2VLM achieves strong performance (PixCorr 0.33, CLIP 85%), suggesting that improvements in brain-to-latent alignment play an important role in reconstruction quality alongside generative modeling. These findings provide empirical evidence for hierarchical alignment between cortical representations and model latent spaces, positioning the brain-to-latent interface as a primary bottleneck in brain decoding systems. Our code can be found at https://github.com/adarsh-crafts/Brain2VLM
- Open Access
- 1 April 2026
- Published by Wiley in International Endodontic Journal
Generative Artificial Intelligence for Computer Vision in Endodontics: A Review of Current State and Future Potential
Background: Computer vision methods based on artificial intelligence (AI) have found numerous applications in endodontic diagnosis and treatment planning. While most current applications employ discriminative deep learning models for detection and classification tasks, the field is now witnessing the rise of generative AI (GenAI), a class of AI models that learn to generate new data samples resembling the original training data. Despite the potential of GenAI, its current state, applications in endodontics and challenges have not been thoroughly explored. Objectives: This narrative review aims to explore the current state and potential applications of GenAI models in computer vision and imaging within endodontics. Methods: A literature search was conducted in PubMed/MEDLINE, Web of Science, Embase, Scopus and arXiv through July 2025, using predefined keywords related to generative models, dental imaging and endodontics. This comprehensive narrative review examines the fundamental principles and current applications of vision GenAI models in endodontics, aiming to clarify their capabilities and evaluate their prospective implications for future applications in endodontic diagnosis, image enhancement and treatment planning. Results: Diverse applications of GenAI in endodontics were identified. Synthetic data generation models can produce unconditional and conditional dental images to augment training datasets and create educational content for rare pathologies. Image enhancement techniques, including super‐resolution, denoising and artefact reduction, show promise in improving diagnostic quality by revealing fine anatomical details. Studies report improvements in quantitative metrics and clinical detection rates, particularly for challenging structures such as second mesiobuccal canals in maxillary molars. Modality conversion applications include 2D‐to‐3D reconstruction from radiographs, cone‐beam computed tomography (CBCT) to magnetic resonance imaging conversion for enhanced soft tissue visualisation, and CBCT to micro‐computed tomography transformation for improved spatial resolution. Treatment enhancement applications encompass customised surgical guide design, treatment outcome prediction and AI‐based computer‐aided design for restoration of endodontically treated teeth. However, current research is predominantly found in other fields of dentistry and emphasises technical feasibility rather than clinical validation in real‐world scenarios. Conclusion: GenAI models are promising tools for enhancing endodontic treatments through data synthesis, image quality improvement and treatment optimisation. However, there are significant gaps between technical demonstrations and clinical implementation.
- 1 April 2026
- Published by Wiley in International Endodontic Journal
Open AccessGenerative Artificial Intelligence for Computer Vision in Endodontics: A Review of Current State and Future Potential
Background: Computer vision methods based on artificial intelligence (AI) have found numerous applications in endodontic diagnosis and treatment planning. While most current applications employ discriminative deep learning models for detection and classification tasks, the field is now witnessing the rise of generative AI (GenAI), a class of AI models that learn to generate new data samples resembling the original training data. Despite the potential of GenAI, its current state, applications in endodontics and challenges have not been thoroughly explored. Objectives: This narrative review aims to explore the current state and potential applications of GenAI models in computer vision and imaging within endodontics. Methods: A literature search was conducted in PubMed/MEDLINE, Web of Science, Embase, Scopus and arXiv through July 2025, using predefined keywords related to generative models, dental imaging and endodontics. This comprehensive narrative review examines the fundamental principles and current applications of vision GenAI models in endodontics, aiming to clarify their capabilities and evaluate their prospective implications for future applications in endodontic diagnosis, image enhancement and treatment planning. Results: Diverse applications of GenAI in endodontics were identified. Synthetic data generation models can produce unconditional and conditional dental images to augment training datasets and create educational content for rare pathologies. Image enhancement techniques, including super‐resolution, denoising and artefact reduction, show promise in improving diagnostic quality by revealing fine anatomical details. Studies report improvements in quantitative metrics and clinical detection rates, particularly for challenging structures such as second mesiobuccal canals in maxillary molars. Modality conversion applications include 2D‐to‐3D reconstruction from radiographs, cone‐beam computed tomography (CBCT) to magnetic resonance imaging conversion for enhanced soft tissue visualisation, and CBCT to micro‐computed tomography transformation for improved spatial resolution. Treatment enhancement applications encompass customised surgical guide design, treatment outcome prediction and AI‐based computer‐aided design for restoration of endodontically treated teeth. However, current research is predominantly found in other fields of dentistry and emphasises technical feasibility rather than clinical validation in real‐world scenarios. Conclusion: GenAI models are promising tools for enhancing endodontic treatments through data synthesis, image quality improvement and treatment optimisation. However, there are significant gaps between technical demonstrations and clinical implementation.
- Top Cited Papers
- 16 January 2026
- Published by Oxford University Press (OUP) in Dentomaxillofacial Radiology
Impact of labelling inaccuracy and image noise on tooth segmentation in panoramic radiographs using federated, centralized, and local learning
Objectives: Federated learning (FL) may mitigate privacy constraints, heterogeneous data quality, and inconsistent labelling in dental artificial intelligence (AI). FL was compared with centralized (CL) and local learning (LL) for tooth segmentation in panoramic radiographs across multiple data corruption scenarios. Methods: An Attention U-Net was trained on 2066 radiographs from 6 institutions across 4 settings: baseline (unaltered data), label manipulation (dilated/missing annotations), image-quality manipulation (additive Gaussian noise), and exclusion of one faulty client with corrupted data. FL was implemented via the Flower AI framework. Per-client training and validation loss trajectories were monitored for anomaly detection and a set of metrics (Dice, IoU, HD, HD95, and ASSD) were evaluated on a hold-out test set. From these metrics significance results were reported through Wilcoxon signed-rank test. CL and LL served as comparators. Results: Baseline: FL achieved a median Dice of 0.949 (ASSD: 1.332), slightly better than CL at 0.947 (ASSD: 1.371) and LL at 0.936-0.940 (ASSD: 1.519-1.698). Label manipulation: FL maintained the best median Dice score at 0.949 (ASSD: 1.465) versus CL’s 0.942 (ASSD: 1.757) and LL’s 0.930-0.940 (ASSD: 1.519-2.115). Similar performance was observed when 2 faulty clients were introduced. Image noise: FL led with a Dice at 0.949 (ASSD: 1.311); CL had a Dice of 0.948 (ASSD: 1.361); LL ranged from 0.932 to 0.940 (ASSD: 1.519-1.774). Similar performance was observed when 2 faulty clients were introduced, with CL performing slightly better than FL. Faulty client exclusion: FL showed a Dice of 0.948 (ASSD: 1.331) better than CL’s 0.946 (ASSD: 1.393). Loss curve monitoring reliably flagged the corrupted site. Conclusions: FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI development.
- 16 January 2026
- Published by Oxford University Press (OUP) in Dentomaxillofacial Radiology
Top Cited PapersImpact of labelling inaccuracy and image noise on tooth segmentation in panoramic radiographs using federated, centralized, and local learning
Objectives: Federated learning (FL) may mitigate privacy constraints, heterogeneous data quality, and inconsistent labelling in dental artificial intelligence (AI). FL was compared with centralized (CL) and local learning (LL) for tooth segmentation in panoramic radiographs across multiple data corruption scenarios. Methods: An Attention U-Net was trained on 2066 radiographs from 6 institutions across 4 settings: baseline (unaltered data), label manipulation (dilated/missing annotations), image-quality manipulation (additive Gaussian noise), and exclusion of one faulty client with corrupted data. FL was implemented via the Flower AI framework. Per-client training and validation loss trajectories were monitored for anomaly detection and a set of metrics (Dice, IoU, HD, HD95, and ASSD) were evaluated on a hold-out test set. From these metrics significance results were reported through Wilcoxon signed-rank test. CL and LL served as comparators. Results: Baseline: FL achieved a median Dice of 0.949 (ASSD: 1.332), slightly better than CL at 0.947 (ASSD: 1.371) and LL at 0.936-0.940 (ASSD: 1.519-1.698). Label manipulation: FL maintained the best median Dice score at 0.949 (ASSD: 1.465) versus CL’s 0.942 (ASSD: 1.757) and LL’s 0.930-0.940 (ASSD: 1.519-2.115). Similar performance was observed when 2 faulty clients were introduced. Image noise: FL led with a Dice at 0.949 (ASSD: 1.311); CL had a Dice of 0.948 (ASSD: 1.361); LL ranged from 0.932 to 0.940 (ASSD: 1.519-1.774). Similar performance was observed when 2 faulty clients were introduced, with CL performing slightly better than FL. Faulty client exclusion: FL showed a Dice of 0.948 (ASSD: 1.331) better than CL’s 0.946 (ASSD: 1.393). Loss curve monitoring reliably flagged the corrupted site. Conclusions: FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI development.
- 1 November 2025
- Published by Springer Nature
- p. 62-72
Is It Large Enough? A Prompt Learning for Large Multi-modal Models
- 1 November 2025
- Published by Springer Nature
- p. 62-72
Is It Large Enough? A Prompt Learning for Large Multi-modal Models
- Open Access
Impact of Labeling Inaccuracy and Image Noise on Tooth Segmentation in Panoramic Radiographs using Federated, Centralized and Local Learning
Objectives: Federated learning (FL) may mitigate privacy constraints, heterogeneous data quality, and inconsistent labeling in dental diagnostic AI. We compared FL with centralized (CL) and local learning (LL) for tooth segmentation in panoramic radiographs across multiple data corruption scenarios. Methods: An Attention U-Net was trained on 2066 radiographs from six institutions across four settings: baseline (unaltered data); label manipulation (dilated/missing annotations); image-quality manipulation (additive Gaussian noise); and exclusion of a faulty client with corrupted data. FL was implemented via the Flower AI framework. Per-client training- and validation-loss trajectories were monitored for anomaly detection and a set of metrics (Dice, IoU, HD, HD95 and ASSD) was evaluated on a hold-out test set. From these metrics significance results were reported through Wilcoxon signed-rank test. CL and LL served as comparators. Results: Baseline: FL achieved a median Dice of 0.94889 (ASSD: 1.33229), slightly better than CL at 0.94706 (ASSD: 1.37074) and LL at 0.93557-0.94026 (ASSD: 1.51910-1.69777). Label manipulation: FL maintained the best median Dice score at 0.94884 (ASSD: 1.46487) versus CL's 0.94183 (ASSD: 1.75738) and LL's 0.93003-0.94026 (ASSD: 1.51910-2.11462). Image noise: FL led with Dice at 0.94853 (ASSD: 1.31088); CL scored 0.94787 (ASSD: 1.36131); LL ranged from 0.93179-0.94026 (ASSD: 1.51910-1.77350). Faulty-client exclusion: FL reached Dice at 0.94790 (ASSD: 1.33113) better than CL's 0.94550 (ASSD: 1.39318). Loss-curve monitoring reliably flagged the corrupted site. Conclusions: FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI deployment.
Open AccessImpact of Labeling Inaccuracy and Image Noise on Tooth Segmentation in Panoramic Radiographs using Federated, Centralized and Local Learning
Objectives: Federated learning (FL) may mitigate privacy constraints, heterogeneous data quality, and inconsistent labeling in dental diagnostic AI. We compared FL with centralized (CL) and local learning (LL) for tooth segmentation in panoramic radiographs across multiple data corruption scenarios. Methods: An Attention U-Net was trained on 2066 radiographs from six institutions across four settings: baseline (unaltered data); label manipulation (dilated/missing annotations); image-quality manipulation (additive Gaussian noise); and exclusion of a faulty client with corrupted data. FL was implemented via the Flower AI framework. Per-client training- and validation-loss trajectories were monitored for anomaly detection and a set of metrics (Dice, IoU, HD, HD95 and ASSD) was evaluated on a hold-out test set. From these metrics significance results were reported through Wilcoxon signed-rank test. CL and LL served as comparators. Results: Baseline: FL achieved a median Dice of 0.94889 (ASSD: 1.33229), slightly better than CL at 0.94706 (ASSD: 1.37074) and LL at 0.93557-0.94026 (ASSD: 1.51910-1.69777). Label manipulation: FL maintained the best median Dice score at 0.94884 (ASSD: 1.46487) versus CL's 0.94183 (ASSD: 1.75738) and LL's 0.93003-0.94026 (ASSD: 1.51910-2.11462). Image noise: FL led with Dice at 0.94853 (ASSD: 1.31088); CL scored 0.94787 (ASSD: 1.36131); LL ranged from 0.93179-0.94026 (ASSD: 1.51910-1.77350). Faulty-client exclusion: FL reached Dice at 0.94790 (ASSD: 1.33113) better than CL's 0.94550 (ASSD: 1.39318). Loss-curve monitoring reliably flagged the corrupted site. Conclusions: FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI deployment.
- Open Access
PanoDiff-SR: Synthesizing Dental Panoramic Radiographs using Diffusion and Super-resolution
There has been increasing interest in the generation of high-quality, realistic synthetic medical images in recent years. Such synthetic datasets can mitigate the scarcity of public datasets for artificial intelligence research, and can also be used for educational purposes. In this paper, we propose a combination of diffusion-based generation (PanoDiff) and Super-Resolution (SR) for generating synthetic dental panoramic radiographs (PRs). The former generates a low-resolution (LR) seed of a PR (256 X 128) which is then processed by the SR model to yield a high-resolution (HR) PR of size 1024 X 512. For SR, we propose a state-of-the-art transformer that learns local-global relationships, resulting in sharper edges and textures. Experimental results demonstrate a Frechet inception distance score of 40.69 between 7243 real and synthetic images (in HR). Inception scores were 2.55, 2.30, 2.90 and 2.98 for real HR, synthetic HR, real LR and synthetic LR images, respectively. Among a diverse group of six clinical experts, all evaluating a mixture of 100 synthetic and 100 real PRs in a time-limited observation, the average accuracy in distinguishing real from synthetic images was 68.5% (with 50% corresponding to random guessing).
Open AccessPanoDiff-SR: Synthesizing Dental Panoramic Radiographs using Diffusion and Super-resolution
There has been increasing interest in the generation of high-quality, realistic synthetic medical images in recent years. Such synthetic datasets can mitigate the scarcity of public datasets for artificial intelligence research, and can also be used for educational purposes. In this paper, we propose a combination of diffusion-based generation (PanoDiff) and Super-Resolution (SR) for generating synthetic dental panoramic radiographs (PRs). The former generates a low-resolution (LR) seed of a PR (256 X 128) which is then processed by the SR model to yield a high-resolution (HR) PR of size 1024 X 512. For SR, we propose a state-of-the-art transformer that learns local-global relationships, resulting in sharper edges and textures. Experimental results demonstrate a Frechet inception distance score of 40.69 between 7243 real and synthetic images (in HR). Inception scores were 2.55, 2.30, 2.90 and 2.98 for real HR, synthetic HR, real LR and synthetic LR images, respectively. Among a diverse group of six clinical experts, all evaluating a mixture of 100 synthetic and 100 real PRs in a time-limited observation, the average accuracy in distinguishing real from synthetic images was 68.5% (with 50% corresponding to random guessing).
- Top Cited PapersOpen Access
- 2 February 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (2) , 41
Pano-GAN: A Deep Generative Model for Panoramic Dental Radiographs
This paper presents the development of a generative adversarial network (GAN) for the generation of synthetic dental panoramic radiographs. While this is an exploratory study, the ultimate aim is to address the scarcity of data in dental research and education. A deep convolutional GAN (DCGAN) with the Wasserstein loss and a gradient penalty (WGAN-GP) was trained on a dataset of 2322 radiographs of varying quality. The focus of this study was on the dentoalveolar part of the radiographs; other structures were cropped out. Significant data cleaning and preprocessing were conducted to standardize the input formats while maintaining anatomical variability. Four candidate models were identified by varying the critic iterations, number of features and the use of denoising prior to training. To assess the quality of the generated images, a clinical expert evaluated a set of generated synthetic radiographs using a ranking system based on visibility and realism, with scores ranging from 1 (very poor) to 5 (excellent). It was found that most generated radiographs showed moderate depictions of dentoalveolar anatomical structures, although they were considerably impaired by artifacts. The mean evaluation scores showed a trade-off between the model trained on non-denoised data, which showed the highest subjective quality for finer structures, such as the mandibular canal and trabecular bone, and one of the models trained on denoised data, which offered better overall image quality, especially in terms of clarity and sharpness and overall realism. These outcomes serve as a foundation for further research into GAN architectures for dental imaging applications.
- 2 February 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (2) , 41
Top Cited PapersOpen AccessPano-GAN: A Deep Generative Model for Panoramic Dental Radiographs
This paper presents the development of a generative adversarial network (GAN) for the generation of synthetic dental panoramic radiographs. While this is an exploratory study, the ultimate aim is to address the scarcity of data in dental research and education. A deep convolutional GAN (DCGAN) with the Wasserstein loss and a gradient penalty (WGAN-GP) was trained on a dataset of 2322 radiographs of varying quality. The focus of this study was on the dentoalveolar part of the radiographs; other structures were cropped out. Significant data cleaning and preprocessing were conducted to standardize the input formats while maintaining anatomical variability. Four candidate models were identified by varying the critic iterations, number of features and the use of denoising prior to training. To assess the quality of the generated images, a clinical expert evaluated a set of generated synthetic radiographs using a ranking system based on visibility and realism, with scores ranging from 1 (very poor) to 5 (excellent). It was found that most generated radiographs showed moderate depictions of dentoalveolar anatomical structures, although they were considerably impaired by artifacts. The mean evaluation scores showed a trade-off between the model trained on non-denoised data, which showed the highest subjective quality for finer structures, such as the mandibular canal and trabecular bone, and one of the models trained on denoised data, which offered better overall image quality, especially in terms of clarity and sharpness and overall realism. These outcomes serve as a foundation for further research into GAN architectures for dental imaging applications.