Benchmarking Deep Learning Models for Tooth Structure Segmentation.

A wide range of deep learning (DL) architectures with varying depths are available, with developers usually choosing one or a few of them for their specific task in a nonsystematic way. Benchmarking (i.e., the systematic comparison of state-of-the art architectures on a specific task) may provide guidance in the model development process and may allow developers to make better decisions. However, comprehensive benchmarking has not been performed in dentistry yet. We aimed to benchmark a range of architecture designs for 1 specific, exemplary case: tooth structure segmentation on dental bitewing radiographs. We built 72 models for tooth structure (enamel, dentin, pulp, fillings, crowns) segmentation by combining 6 different DL network architectures (U-Net, U-Net++, Feature Pyramid Networks, LinkNet, Pyramid Scene Parsing Network, Mask Attention Network) with 12 encoders from 3 different encoder families (ResNet, VGG, DenseNet) of varying depth (e.g., VGG13, VGG16, VGG19). On each model design, 3 initialization strategies (ImageNet, CheXpert, random initialization) were applied, resulting overall into 216 trained models, which were trained up to 200 epochs with the Adam optimizer (learning rate = 0.0001) and a batch size of 32. Our data set consisted of 1,625 human-annotated dental bitewing radiographs. We used a 5-fold cross-validation scheme and quantified model performances primarily by the F1-score. Initialization with ImageNet or CheXpert weights significantly outperformed random initialization (P < 0.05). Deeper and more complex models did not necessarily perform better than less complex alternatives. VGG-based models were more robust across model configurations, while more complex models (e.g., from the ResNet family) achieved peak performances. In conclusion, initializing models with pretrained weights may be recommended when training models for dental radiographic analysis. Less complex model architectures may be competitive alternatives if computational resources and training time are restricting factors. Models developed and found superior on nondental data sets may not show this behavior for dental domain-specific tasks.

Perivascular vital cells in the ablation center after multibipolar radiofrequency ablation in an in vivo porcine model.

Multibipolar radiofrequency ablation (RFA) is an advanced ablation technique for early stage hepatocellular carcinoma and liver metastases. Vessel cooling in multibipolar RFA has not been systematically investigated. The objective of this study was to evaluate the presence of perivascular vital cells within the ablation zone after multibipolar RFA. Multibipolar RFA were performed in domestic pigs in vivo. Three internally cooled bipolar RFA applicators were used simultaneously. Three experimental settings were planned: (1) inter-applicator-distance: 15 mm; (2) inter-applicator-distance: 20 mm; (3) inter-applicator-distance: 20 mm with hepatic inflow occlusion (Pringle maneuver). A vitality staining was used to analyze liver cell vitality around all vessels in the ablation center with a diameter > 0.5 mm histologically. 771 vessels were identified. No vital tissue was seen around 423 out of 429 vessels (98.6%) situated within the central white zone. Vital cells could be observed around major hepatic vessels situated adjacent to the ablation center. Vessel diameter (> 3.0 mm; p < 0.05) and low vessel-to-ablation-center distance (< 0.2 mm; p < 0.05) were identified as risk factors for incomplete ablation adjacent to hepatic vessels. The vast majority of vessels, which were localized in the clinically relevant white zone, showed no vital perivascular cells, regardless of vessel diameter and vessel type. However, there was a risk of incomplete ablation around major hepatic vessels situated directly within the ablation center. A Pringle maneuver could avoid incomplete ablations.

Influence of interapplicator distance on multibipolar radiofrequency ablation during physiological and interrupted liver perfusion in an in vivo porcine model.

Radiofrequency ablation (RFA) is a curative treatment option for early stage hepatocellular carcinoma (HCC). Vascular inflow occlusion to the liver (Pringle manoeuvre) and multibipolar RFA (mbRFA) represent possibilities to generate large ablations. This study evaluated the impact of different interapplicator distances and a Pringle manoeuvre on ablation area and geometry of mbRFA. 24 mbRFA were planned in porcine livers in vivo. Test series with continuous blood flow had an interapplicator distance of 20 mm and 15 mm, respectively. For a Pringle manoeuvre, interapplicator distance was predefined at 20 mm. After liver dissection, ablation area and geometry were analysed macroscopically and histologically. Confluent and homogenous ablations could be achieved with a Pringle manoeuvre and an interapplicator distance of 15 mm with sustained hepatic blood flow. Ablation geometry was inhomogeneous with an applicator distance of 20 mm with physiological liver perfusion. A Pringle manoeuvre led to a fourfold increase in ablation area in comparison to sustained hepatic blood flow (p < 0.001). Interapplicator distance affects ablation geometry of mbRFA. Strict adherence to the planned applicator distance is advisable under continuous blood flow. The application of a Pringle manoeuvre should be considered when compliance with the interapplicator distance cannot be guaranteed.