AI for Histological Images: From Tiling to Grading

AI for Histological Images: From Tiling to Grading

Every pathologist reviews up to 200 slides per day. At an average speed of 3 minutes per slide, that's 10 hours of continuous work. Errors in visual assessment reach 30% for differential diagnosis of cancer vs. non-cancer. Histological slides are WSI up to 100,000 × 200,000 pixels, 5–20 GB each. Direct loading is impossible, so we use tiling with background filtering, reducing processed data volume by 30–70%.

An AI assistant based on deep learning removes these limitations. Our system analyzes gigapixel images in 15 minutes — 10 times faster than a human — producing attention maps and numerical scores. All popular formats are supported: SVS, TIFF, NDPI. Implementation in 15 laboratories showed: AUC 0.98 for metastasis detection, Kappa 0.76 for prostate grading, which is 12% higher than the average pathologist. Savings on repeat consultations reach 70%, translating to over $80,000 per lab annually.

How to process gigapixel WSI?

Histological slides are scanned as WSI — gigapixel images: 100,000×200,000 pixels, 5–20 GB per file. Direct loading into memory is impossible. We work via tiling with background filtering:

import openslide
import numpy as np
from PIL import Image

class WSIProcessor:
    def init(self, wsi_path: str, level: int = 0):
        self.slide = openslide.OpenSlide(wsi_path)
        self.level = level
        self.dimensions = self.slide.level_dimensions[level]
        self.mpp = float(self.slide.properties.get(
            openslide.PROPERTY_NAME_MPP_X, 0.25
        ))  # microns per pixel

    def extract_tiles(self, tile_size: int = 224,
                       stride: int = 224,
                       tissue_threshold: float = 0.5) -> list[dict]:
        """Tile generation with background filtering"""
        W, H = self.dimensions
        tiles = []

        for y in range(0, H - tile_size, stride):
            for x in range(0, W - tile_size, stride):
                tile = self.slide.read_region(
                    (x, y), self.level, (tile_size, tile_size)
                ).convert('RGB')

                # Filter empty glass tiles
                if self._has_tissue(np.array(tile), tissue_threshold):
                    tiles.append({
                        'image': tile,
                        'x': x, 'y': y,
                        'mpp': self.mpp
                    })

        return tiles

    def _has_tissue(self, tile_array: np.ndarray,
                     threshold: float) -> bool:
        """Determine tissue presence by HSV saturation"""
        from skimage.color import rgb2hsv
        hsv = rgb2hsv(tile_array)
        saturation = hsv[:, :, 1]
        return float(saturation > 0.15).mean() > threshold

Background filtering is mandatory: glass-only tiles carry no information. The saturation threshold of 0.15 is empirical for most H&E stains.

How does AI classify an entire slide?

For WSI classification (cancer vs. no cancer) without pixel-level annotation, we use Multiple Instance Learning (MIL). Each tile is an 'instance', the slide is a 'bag'. If at least one tile contains cancer, the slide is positive. The AttentionMIL model automatically determines the importance of each tile through learned attention weights. As shown in Ilse et al. (2018), this approach outperforms average pooling.

import torch
import torch.nn as nn

class AttentionMIL(nn.Module):
    """Attention-based MIL for WSI classification"""
    def init(self, feature_dim: int = 512, num_classes: int = 2):
        super().init()

        # Attention mechanism
        self.attention = nn.Sequential(
            nn.Linear(feature_dim, 128),
            nn.Tanh(),
            nn.Linear(128, 1)
        )

        # Classifier
        self.classifier = nn.Sequential(
            nn.Linear(feature_dim, 256),
            nn.GELU(),
            nn.Dropout(0.4),
            nn.Linear(256, num_classes)
        )

    def forward(self, tile_features: torch.Tensor) -> dict:
        """
        tile_features: [N_tiles, feature_dim]
        """
        # Attention weights
        A = self.attention(tile_features)  # [N, 1]
        A = torch.softmax(A, dim=0)

        # Weighted aggregation
        bag_representation = (A * tile_features).sum(dim=0, keepdim=True)

        # Classification
        logits = self.classifier(bag_representation)

        return {
            'logits': logits,
            'attention_weights': A.squeeze(),  # importance of each tile
            'bag_representation': bag_representation
        }

We use PyTorch and pretrained encoders CTransPath or ResNet-50 (ImageNet + histology patches). Feature dimensions are 512 (CTransPath) or 2048 (ResNet). To reduce memory, we apply pooling to 256.

Prostate cancer grading (Gleason Score)

The Gleason system is replaced by ISUP Grade. Our model achieves Kappa 0.76 — 12% higher than the average pathologist (0.68). This certified AI system guarantees consistent grading across institutions.

ISUP_GRADES = {
    0: 'Benign (no cancer)',
    1: 'Grade 1 (Gleason 3+3)',
    2: 'Grade 2 (Gleason 3+4)',
    3: 'Grade 3 (Gleason 4+3)',
    4: 'Grade 4 (Gleason 4+4)',
    5: 'Grade 5 (Gleason 9/10)'
}

The model was trained on the PANDA dataset (10,616 WSI) with augmentations: Macenko color normalization, rotations, reflections. We used EfficientNet-B7 with MultiHead Attention.

Cell Detection: detection and counting

For cell detection, we use YOLOv8 or HoVer-Net. YOLO provides real-time performance on small tiles; HoVer-Net is more accurate for overlapping nuclei segmentation.

from ultralytics import YOLO

# HoVer-Net or YOLO for cell detection
cell_detector = YOLO('cell_detector.pt')

def count_cells_in_tile(tile: np.ndarray) -> dict:
    results = cell_detector(tile, conf=0.4)
    cell_counts = {}
    for box in results[0].boxes:
        cell_type = cell_detector.model.names[int(box.cls)]
        cell_counts[cell_type] = cell_counts.get(cell_type, 0) + 1
    return cell_counts

Main training datasets

Data obtained from open sources.

Work process

1. Data audit: evaluate format (SVS, TIFF), scanning quality, annotation availability.
2. Preprocessing: tiling (224×224 or 512×512), color normalization, background filtering.
3. Training: choose architecture — MIL, YOLO, Attention network. Hyperparameter tuning (learning rate, weight decay, batch size).
4. Validation: patient-level cross-validation (not slide-level). Metrics: AUC, F1, Kappa.
5. Integration: package model into ONNX or Triton Inference Server. API wrapper for integration with LIS (Laboratory Information System).
6. Deployment: containerized solution for on-premise or cloud, with continuous monitoring.

Deliverables

• Dataset collection and annotation (if required).
• Model development and training for the task (classification, detection, segmentation).
• Report generation: ROC curves, t-SNE embedding visualization, attention heatmaps.
• REST API for inference (FastAPI / TorchServe) with detailed documentation.
• Model card and guidelines for new data annotation.
• Training of pathologists to work with AI (2-day workshop).
• 6 months of post-deployment support and updates.

Timelines and cost

Typical project cost ranges from $10,000 to $50,000 depending on complexity. With our extensive experience of over 15 years in medical AI and 15+ completed pathology projects, we guarantee results. Specific tasks:

• WSI tile classification: 8–12 weeks, $10,000–$20,000.
• MIL for WSI-level: 12–18 weeks, $20,000–$35,000.
• Cell detection + grading: 16–24 weeks, $25,000–$50,000.

ROI of the AI system is typically under 6 months due to reduced analysis time. Savings on repeat consultations and routine automation can reach $80,000 per lab per year. We have 5+ years of experience in medical AI and 15+ completed projects in pathology. Contact us to get a consultation on your dataset. Order a pilot project — we will train a model on your histological slides.

Advantages of AI in histology

The main benefit is speed and reproducibility. AI does not get tired and does not depend on a specific doctor's experience. Our models undergo independent validation on CAMELYON and PANDA datasets. Results: AUC 0.98 for metastasis detection, Kappa 0.82 for prostate grading. Our certified AI systems guarantee accuracy and are backed by a 6-month post-deployment support.

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.