Urolithiasis detection based on data fusion of laser-induced breakdown spectroscopy and fourier transform infrared spectroscopy.

Most urolithiasis are only diagnosed after symptoms appear, which exacerbates patient suffering, and in severe cases can cause irreparable damage. Thus, there is an urgent need for simple, rapid, and cost-effective detection methods to early detection of urolithiasis. To address the issue, This work proposes a multi-scale driven channel-attention fusion method (MCANet) for dual-spectral data. This approach extracts complementary urine information: elemental data via laser-induced breakdown spectroscopy (LIBS) and detailed molecular data via fourier transform infrared spectroscopy (FTIR), then efficiently fuses them. Using the inherent correlation and complementarity between atomic and molecular features, MCANet enables urolithiasis detection. The results demonstrate that MCANet achieves an Accuracy of 96.36%, Precision of 95.13%, Recall of 97.72%, F1-score of 96.44%, and AUC of 0.98. The model consistently outperforms single-spectrum models by more than 9.09%, and exceed SVM and ResNet18 by 7.41%. This model provides a novel, rapid and accurate method for early detection of urolithiasis in the future.