TPATH Glomeruli Detection - YOLOv8 Model

Automated detection and classification of glomeruli in renal pathology whole slide images.

RESEARCH USE ONLY - This model is intended for research purposes only and has not been validated for clinical diagnostic use.

Model Description

Model Type: YOLOv8 Object Detection
Input Size: 1024×1024 pixels
Classes: 2 (normal glomeruli, sclerotic glomeruli)
Framework: Ultralytics YOLOv8

Quick Start

Installation

pip install ultralytics huggingface_hub

Download Model

# Using Hugging Face CLI
hf download shuheimisawa/tpath-glomeruli-v1 best.pt --local-dir ./models/

# Or using Python
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(repo_id="shuheimisawa/tpath-glomeruli-v1", filename="best.pt")

Basic Usage

from ultralytics import YOLO

# Load model
model = YOLO("models/best.pt")

# Run inference on an image
results = model("path/to/kidney_tissue.png", conf=0.20)

# Process results
for result in results:
    boxes = result.boxes
    for box in boxes:
        class_id = int(box.cls[0])
        confidence = float(box.conf[0])
        class_name = "normal" if class_id == 0 else "sclerotic"
        print(f"Detected {class_name} glomerulus with confidence {confidence:.2f}")

Whole Slide Image Processing

For processing whole slide images (.svs, .ndpi, .tiff), use the full pipeline:

git clone https://github.com/shuheimisawa/glomeruli-detector.git
cd glomeruli-detector
pip install -r requirements.txt

from tpath_glomeruli import AggressiveOverlapMerge

# Initialize detector with Test-Time Augmentation
detector = AggressiveOverlapMerge(
    model_path="models/best.pt",
    confidence_threshold=0.20,
    overlap_threshold=0.20,
    use_tta=True  # Improves detection accuracy
)

# Process whole slide image
results = detector.process_wsi(
    wsi_path="path/to/slide.svs",
    output_dir="results"
)

print(f"Total glomeruli: {results['total_detections']}")
print(f"Normal: {results['normal_count']}")
print(f"Sclerotic: {results['sclerotic_count']}")

Output Format

Detection results include:

Bounding box coordinates (x_min, y_min, x_max, y_max)
Center coordinates (center_x, center_y)
Class name (normal/sclerotic)
Confidence score (0-1)

Recommended Settings

Parameter	Value	Description
Confidence Threshold	0.20	Minimum detection confidence
Overlap Threshold	0.20	IoU threshold for merging duplicates
Use TTA	True	Test-Time Augmentation for better recall

Supported Formats

Aperio SVS (.svs)
TIFF (.tif, .tiff)
Hamamatsu NDPI (.ndpi)
Leica SCN (.scn)
Standard images (.png, .jpg)

Limitations

Trained on specific staining protocols; performance may vary on different tissue preparations
Processing time: ~30-60 seconds per slide depending on size
Requires GPU for optimal performance (CPU inference is slower)

License

MIT License

Disclaimer

This model is provided for research purposes only. It has not been validated for clinical use and should not be used for medical diagnosis or treatment decisions. Always consult qualified medical professionals for clinical applications.

Contact

Issues: GitHub Issues
Email: [email protected]

Downloads last month: 25

shuheimisawa
/

tpath-glomeruli-v1