AI Analysis of Dental X-Rays: Automating Interpretation

AI Analysis of Dental X-Rays: Automating Interpretation

A radiologist spends an average of 5–7 minutes describing one orthopantomogram. With a flow of 30+ images per day, attention inevitably wanes, especially in low-contrast areas (periapical lesions, initial caries). A practical case: after implementing our system in one clinic, the number of missed lesions smaller than 2 mm decreased by 60%, and the average time per image dropped to 40 seconds. We develop AI systems that take over routine annotation — tooth numbering, pathology detection, and bone tissue assessment. This is not a replacement for the doctor, but a second opinion in seconds, with flags for suspicious zones and prioritization.

We offer a turnkey solution: from dataset collection and annotation to integration into your DICOM infrastructure. We'll assess your project for free — just write to us.

Main Problems the System Solves

The main pain points the system addresses: missing small pathologies (lesions less than 2 mm on OPG are often invisible to the naked eye), non-standardized image quality (low exposure, motion artifacts, varying contrast — models with augmentation adapt to any equipment), and doctor fatigue (after the 20th image, accuracy drops; the system does not tire and delivers stable results with p99 latency < 500 ms). Additionally, the system prioritizes images with a high probability of pathology, allowing the doctor to focus on complex cases.

Types of Dental X-Rays and Tasks

How Tooth Numbering and Pathology Detection on OPG Works

This is the central task for panoramic images. Teeth are numbered according to the FDI system (11–48) — two levels of classification: quadrant + position. The model must detect each tooth as a separate object with a class = FDI number.

Specifics: teeth partially overlap (especially lower molars + mandibular canal), object sizes vary by 3–4 times, radiographic density depends on the machine and exposure — a non-standardized input.

Solution: Instance segmentation via Mask R-CNN or YOLOv8-seg with pretrained weights on dental datasets (e.g., Tufts Dental Database). Augmentation: brightness/contrast variation ±30%, horizontal flip (jaw symmetry), elastic deformation to simulate anatomical variation. Metrics on the test set: [email protected] = 0.84 for tooth numbering, precision 0.79 for caries cavity detection.

Analysis of Periapical Changes (PAI)

Periapical changes — darkening at the root apex, a sign of inflammation. Lesion size: from 1–2 mm to 10+ mm. Small lesions on OPG are easily missed visually. Task: regression + classification. A segmentation model outputs a lesion mask; a separate classifier evaluates the PAI score (1–5).

Challenge: normal anatomy (mental foramen, sinuses) visually resembles pathology. False-positive predictions irritate the doctor and reduce trust in the system. Solution — two-stage verification: if the model flags a zone with confidence 0.4–0.7 — label it as "requires attention" (not "pathology"). At confidence > 0.7 — a direct flag.

# Example of two-stage classification with confidence thresholds
def classify_periapical(prediction_logits, threshold_flag=0.7,
                         threshold_watch=0.4):
    probs = torch.sigmoid(prediction_logits)
    labels = torch.zeros_like(probs, dtype=torch.long)
    labels[probs > threshold_watch] = 1   # requires attention
    labels[probs > threshold_flag] = 2    # pathology
    return labels  # 0=normal, 1=watch, 2=flag

How Integration with DICOM and Dental Software Works

Images are stored in DICOM format (.dcm). For reading — pydicom, pixel normalization — custom windowing. Pixel spacing from the DICOM header is needed for calibrating linear measurements (lesion size in mm, root length). Integration with dental practice management systems (Romexis, CS Imaging, Planmeca) via HL7/FHIR API or export of annotations in DICOM SR (Structured Reporting). More about the DICOM format can be found in the official documentation.

What Turnkey AI System Development Includes

• Dataset collection and annotation (minimum 5000 images per task).
• Architecture selection (YOLOv8, U-Net, 3D U-Net) considering your requirements for latency and accuracy.
• Training with validation on your own data.
• Deployment in your infrastructure (Docker, Triton Inference Server).
• Documentation and staff training.
• Warranty of module operation for 6 months after delivery.

Timeline and Cost

Basic module (numbering + caries detection): 8–12 weeks. Full system (OPG + periapical + bitewings + reporting): 16–24 weeks. Cost is calculated considering medical certification requirements and data volume. For reference, implementation typically pays for itself in 6–8 months by reducing manual work. Contact us — we'll assess the project in 1 day.

Why Choose Us

• 5+ years of experience in medical CV.
• 30+ implemented projects in medical image analysis.
• Proprietary datasets and pretrained models for a quick start.
• Post-implementation support: we update models, add new image types on request.

According to the Tufts Dental Database, the average OPG analysis time is reduced by 70% using our system.

Order a pilot module or get a consultation — we'll help you assess the automation potential in your diagnostic center.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.