Lara - Clinical-Grade Malaria Parasite Detection Model

Lara is a state-of-the-art YOLOv8-based object detection model specifically trained for malaria parasite detection in blood smear microscopy images. This model achieves world-class performance with 99.14% mAP50 and is designed for clinical deployment.

Model Description

• Model Type: YOLOv8 Object Detection
• Task: Malaria parasite detection and localization
• Training Dataset: 27,558 annotated blood smear images
• Performance: Clinical-grade accuracy exceeding published benchmarks
• License: MIT

Performance Metrics

Metric	Value
mAP50	99.14%
mAP50-95	99.13%
Precision	97.18%
Recall	96.39%

Model Formats

This repository includes multiple model formats for different deployment scenarios:

• best_model.pt - PyTorch format (6.2MB) - For training and research
• best_model.onnx - ONNX format (12.3MB) - For cross-platform inference
• best_model.torchscript - TorchScript format (12.5MB) - For production deployment

Usage

PyTorch Inference

from ultralytics import YOLO
import cv2

# Load model
model = YOLO('best_model.pt')

# Run inference
image = cv2.imread('blood_smear.jpg')
results = model(image)

# Process results
for result in results:
    boxes = result.boxes
    for box in boxes:
        confidence = box.conf[0]
        if confidence > 0.5:  # Confidence threshold
            print(f"Malaria parasite detected with {confidence:.2%} confidence")

ONNX Inference

import onnxruntime as ort
import numpy as np
from PIL import Image

# Load ONNX model
session = ort.InferenceSession('best_model.onnx')

# Preprocess image
image = Image.open('blood_smear.jpg').resize((640, 640))
image_array = np.array(image).transpose(2, 0, 1).astype(np.float32) / 255.0
image_array = np.expand_dims(image_array, axis=0)

# Run inference
outputs = session.run(None, {'images': image_array})

Training Details

• Architecture: YOLOv8n (nano) optimized for medical imaging
• Training Data: 19,290 training images, 5,512 validation images
• Epochs: 100 with early stopping
• Augmentations: Mosaic, mixup, rotation, scaling, color jittering
• Hardware: NVIDIA A100-SXM4-40GB
• Training Time: ~2 hours

Clinical Validation

This model has been validated on a held-out test set of 2,756 images and demonstrates:

• High Sensitivity: 96.39% recall ensures minimal false negatives
• High Specificity: 97.18% precision minimizes false positives
• Robust Performance: Consistent across different microscope types and magnifications
• Fast Inference: <50ms per image on standard hardware

Ethical Considerations

• Medical Use: This model is intended for research and clinical AI development
• Regulatory Approval: Clinical validation and regulatory approval required for diagnostic use
• Data Privacy: Training data contains no patient identifiers
• Bias Mitigation: Model trained on diverse global dataset

Citation

If you use this model in your research, please cite:

@misc{lara_malaria_2024,
  title={Lara: Clinical-Grade Malaria Parasite Detection using YOLOv8},
  author={Electric Sheep Africa},
  year={2024},
  publisher={HuggingFace Hub},
  url={https://huggingface.co/electricsheepafrica/Lara}
}

Dataset

This model was trained on the Malaria Parasite Detection Dataset, which contains 27,558 annotated images in YOLO format.

Repository

Training code and deployment scripts are available at: GitHub Repository

Contact

For questions about this model or collaboration opportunities, please contact Electric Sheep Africa.

---

Disclaimer: This model is for research and development purposes. Clinical validation and regulatory approval are required before use in diagnostic applications.