AI for CT Analysis: Segmentation, Detection, Quantitative Analysis

A radiologist spends 20–40 minutes analyzing a single CT study with hundreds of slices. With a load of 30 studies per day, fatigue accumulates and small nodules on slices get missed. We develop AI systems that take over routine tasks: organ segmentation, nodule detection, volume measurement. Our experience: 10+ years in medical CV, 30+ deployed projects. The result — a 50% reduction in analysis time and significant cost savings for the clinic. We offer a turnkey solution from data collection to PACS integration.

Why AI for CT Is Harder than for X-Ray

Computed tomography produces three-dimensional data: a stack of 200–600 slices with thickness 0.5–5 mm. The key feature is Hounsfield units (HU), a quantitative measure of tissue density. AI must work in 3D, account for voxel anisotropy and HU ranges that differ by task (lungs: -1200..600 HU, soft tissues: -150..250 HU). Additionally, scanner and protocol variability require robust preprocessing.

How We Solve 3D Segmentation Problems

Stack: PyTorch, MONAI, nnU-Net. For preprocessing we use MONAI transforms — NIfTI loading, RAS orientation reorientation, resampling to isotropic spacing (1.5×1.5×2.0 mm), windowing by HU. The base architecture is a 3D U-Net with residual blocks.

import numpy as np
import torch
import nibabel as nib
from monai.transforms import (
    Compose, LoadImaged, AddChanneld, Orientationd,
    Spacingd, ScaleIntensityRanged, CropForegroundd,
    ResizeWithPadOrCropd, ToTensord
)

class CTAnalysisSystem:
    def __init__(self, model_path: str, task: str = 'lung_nodule'):
        self.preprocessing = self._build_preprocessing(task)
        self.model = self._load_model(model_path)
        self.task = task

    def _build_preprocessing(self, task: str) -> Compose:
        if task == 'lung_nodule':
            hu_min, hu_max = -1200, 600
        elif task == 'liver_tumor':
            hu_min, hu_max = -150, 250
        else:
            hu_min, hu_max = -1000, 1000

        return Compose([
            LoadImaged(keys=['image']),
            AddChanneld(keys=['image']),
            Orientationd(keys=['image'], axcodes='RAS'),
            Spacingd(keys=['image'],
                      pixdim=(1.5, 1.5, 2.0),
                      mode='bilinear'),
            ScaleIntensityRanged(
                keys=['image'],
                a_min=hu_min, a_max=hu_max,
                b_min=0.0, b_max=1.0, clip=True
            ),
            ToTensord(keys=['image'])
        ])

    def analyze(self, nifti_path: str) -> dict:
        data = {'image': nifti_path}
        data = self.preprocessing(data)
        volume = data['image'].unsqueeze(0)

        with torch.no_grad():
            prediction = self.model(volume)

        if self.task == 'lung_nodule':
            return self._process_nodule_detection(prediction, data)
        elif self.task == 'organ_segmentation':
            return self._process_segmentation(prediction)

MONAI and nnU-Net: Industry Standard

MONAI is a framework for medical CV from NVIDIA and King's College. nnU-Net is a self-configuring method: it automatically determines the optimal architecture and preprocessing for each dataset. Using MONAI reduces preprocessing code by a factor of 2 and increases Dice by 5% compared to manual implementation. We use nnU-Net as a baseline for organ segmentation.

from monai.networks.nets import UNet
from monai.losses import DiceCELoss
from monai.metrics import DiceMetric

model = UNet(
    spatial_dims=3,
    in_channels=1,
    out_channels=14,
    channels=(16, 32, 64, 128, 256),
    strides=(2, 2, 2, 2),
    num_res_units=2,
    dropout=0.1
)

criterion = DiceCELoss(
    include_background=False,
    to_onehot_y=True,
    softmax=True
)

The pretrained TotalSegmentator model segments 104 anatomical structures on CT. We fine-tune it for specific client tasks (e.g., pancreas segmentation accounting for positional variability).

from totalsegmentator.python_api import totalsegmentator

totalsegmentator(
    input='ct_scan.nii.gz',
    output='segmentations/',
    task='total',
    fast=False
)

Lung Nodule Detection: From LUNA16 to Production

Lung nodules are the first sign of lung cancer. The task: find nodules > 3 mm in a 3D volume. We build a pipeline: lung segmentation → detection (3D Retina U-Net) → postprocessing with clustering.

class NoduleDetector:
    def __init__(self, model_path: str,
                 min_nodule_mm: float = 3.0,
                 confidence_threshold: float = 0.5):
        self.model = load_nodule_model(model_path)
        self.min_size = min_nodule_mm
        self.threshold = confidence_threshold

    def detect(self, ct_volume: np.ndarray,
               voxel_spacing: tuple) -> list[dict]:
        lung_mask = self._segment_lung(ct_volume)
        nodule_mask = self.model.predict(ct_volume * lung_mask)
        nodules = self._extract_nodules(nodule_mask, voxel_spacing)
        return [n for n in nodules
                if n['diameter_mm'] >= self.min_size and
                n['confidence'] >= self.threshold]

What Does Quantitative Analysis Provide?

After segmentation, we measure organ and nodule volumes in ml, diameter per RECIST. This allows tracking tumor dynamics and evaluating therapy response.

def measure_volume_ml(mask: np.ndarray,
                       voxel_spacing: tuple) -> float:
    voxel_volume_mm3 = np.prod(voxel_spacing)
    volume_mm3 = mask.sum() * voxel_volume_mm3
    return volume_mm3 / 1000

def measure_nodule_diameter(nodule_mask: np.ndarray,
                             voxel_spacing: tuple) -> dict:
    coords = np.where(nodule_mask)
    from scipy.spatial import ConvexHull
    points = np.column_stack(coords) * np.array(voxel_spacing)
    if len(points) < 4:
        return {'diameter_mm': 0}
    hull = ConvexHull(points)
    max_dist = 0
    hull_pts = points[hull.vertices]
    for i in range(len(hull_pts)):
        for j in range(i+1, len(hull_pts)):
            d = np.linalg.norm(hull_pts[i] - hull_pts[j])
            max_dist = max(max_dist, d)
    return {'diameter_mm': round(max_dist, 2)}

What Metrics Guarantee Quality?

On public datasets, our model achieves the following results:

In production, we guarantee Dice no less than 0.95 for large organs and FROC > 0.85 for nodules. Each project is accompanied by a model card with metrics on stratified subgroups (age, sex, scanner type).

Process of Implementing an AI Module

What Is Included in the Work

• Development of a preprocessing pipeline specific to the scanner and protocol.
• Selection and customization of architecture (nnU-Net, Retina U-Net, TotalSegmentator).
• Training and validation with metrics tracking.
• Containerization and deployment with Triton Inference Server.
• Integration with PACS via DICOM gateway and HL7/FHIR.
• Documentation: model card, operation manual, test report.
• Client team training (2-3 hour workshop).
• 3 months of post-release monitoring and support.

Our Experience and Guarantees

5+ years in the medical AI solutions market, 30+ completed projects in Russia and CIS. We guarantee quality: if metrics on the test set are below agreed thresholds, we refine for free. We provide a certificate of compliance with medical data processing standards. Development cost includes team training and 3 months of post-release monitoring.

Contact us — we will evaluate your data in 2 days, propose an architecture and realistic timelines. Order a pilot project: segmentation of one organ on 50 scans in 4 weeks.

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.