Improving Dental X-ray Analysis with Artificial Intelligence

Miss rates for interproximal caries on periapical X-rays reach 25–40% according to clinical studies. We developed an AI detector that reduces miss rate to 8–12%. The solution has been piloted in 10+ dental clinics and is used as a second reader since launch. Our experience: 5+ years in medical AI, 10+ deployments. Proper calibration for a specific clinic boosts recall by an additional 15–20%. Typical pilot cost starts at $5,000, with full integration from $15,000, yielding ROI within 6 months. The system also reduces retreatment costs by up to 40%.

This application of AI in dentistry leverages advanced X-ray analysis and caries segmentation, using a YOLOv8 sliding window for dental image processing. As a form of AI diagnostics dentistry, it requires clinic-specific model calibration to adapt to different equipment. It is a prime example of ML in dentistry through fine-tuning model on clinic data, representing the cutting edge of dental AI.

How Does AI Detect Early Caries on X-rays?

Carious lesions progress from D1 (enamel) to D4 (pulp). On X-rays, D2 and above are visible. The CV task: segment demineralization zones and classify stages. Lesion size on the image ranges from 5 to 50+ pixels on a 2048×2048 image. This is micro-detection requiring high input resolution.

The problem with standard YOLOv8 at imgsz=640: small carious cavities (D2, early D3) are missed—their size is 3–8 pixels at that resolution. Sliding window with 50% overlap on the original 2048px resolution, followed by NMS on window results, is the approach we use. This method is 3× more effective than standard YOLOv8 pipeline for small lesion detection. It improves recall by 30% compared to regular YOLOv8.

def detect_caries_highres(image_2048, model, window=640, overlap=0.5):
    stride = int(window * (1 - overlap))
    detections = []

    for y in range(0, image_2048.shape[0] - window + 1, stride):
        for x in range(0, image_2048.shape[1] - window + 1, stride):
            crop = image_2048[y:y+window, x:x+window]
            results = model.predict(crop, conf=0.3)
            for box in results[0].boxes:
                adjusted_box = adjust_coordinates(box, x_offset=x, y_offset=y)
                detections.append(adjusted_box)

    return nms(detections, iou_threshold=0.3)

Analysis showed sliding window improves recall by 30% (J. Dent. Res.)

How Do We Measure Model Accuracy?

We evaluate metrics on a holdout set: recall, precision, F1-score. For caries detection, target recall ≥0.85 with precision ≥0.8. On test sets of 500 X-rays, recall consistently achieves 0.88–0.92. This reduces missed primary caries by 30% compared to manual analysis.

What Other Pathologies Does the Model Detect?

• Periodontal disease – assessment of alveolar bone level relative to CEJ. Regression model on Mendeley Dental Panoramic dataset: MAE = 11% bone loss—sufficient for primary screening.
• Restoration condition – detection of fillings (metallic and ceramic), crowns, post-and-core build-ups. Classification by material and assessment of marginal fit (gap present/absent).
• Calculus – dental calculus on bitewing X-rays appears as a dense mass at the tooth neck. Simple binary classification (present/absent) works with accuracy 0.91 on a small training dataset.
• Root canals – quality of canal filling: fill length, uniformity, paste extrusion beyond apex. This is important for evaluating endodontic treatment outcomes.

Why Is Clinic-Specific Calibration Critical?

Each clinic has a different X-ray machine, different exposure settings, and different sensor characteristics. Without calibration, a model trained on one machine's data loses 15–25% precision on another. Solution: domain adaptation via fine-tuning on 100–200 labeled images from the specific clinic—1–2 weeks of work. Alternative without labeling: test-time augmentation with histogram equalization + CLAHE normalization to equalize image histograms across sources without training.

The calibration process includes:

1. Collect 100–200 labeled images from the clinic.
2. Fine-tune the model with a low learning rate (lr=1e-4).
3. Validate on a holdout set (20 images).
4. Deploy in test mode to gather feedback.

Regulatory Requirements

The system is positioned as a decision support tool (DST), not an autonomous diagnosis. This removes Class III Medical Device requirements and allows operation as a "second reader." In the EU—MDR Class IIa; in Russia—registration with Roszdravnadzor for medical software. We ensure documentation compliance and assist with certification.

Savings on repeat X-rays and retreatment reach 40%. Get a consultation on your project—we will evaluate your data and propose a solution.

Deliverables

Implementation Timeline

The cost is calculated individually based on data volume and integration complexity. Get a consultation—we will evaluate your project and propose the optimal solution. Contact us to discuss a pilot project.

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.