huzie111/lung-ultrasound-ai

Lung Ultrasound AI Model

Model Description

Multi-task model for lung ultrasound analysis:

• Classification: COVID-19, Other Lung Disease, Healthy
• Segmentation: B-line detection and localization

Architecture

• Encoder: EfficientNet-B3
• Decoder: SegFormer-style MLP
• Multi-task: Hard parameter sharing

Training Data

• Classification: 1,062 images from Ugandan hospitals
• Segmentation: 401 images with B-line annotations

Performance

Metric	Value
Accuracy	23.40%
Precision	39.96%
Model Size	44 MB

Usage

import torch
from model import LungUltrasoundModel

model = LungUltrasoundModel()
model.load_state_dict(torch.load('pytorch_model.bin'))
model.eval()

# Preprocess image (224x224 RGB, normalized to 0-1)
# Run inference
class_logits, seg_logits = model(image_tensor)