Early screening for echinococcosis is crucial, yet current methods have limitations. Raman spectroscopy offers fast, nondestructive detection but suffers from low signal intensity, fluorescence interference, and environmental noise. This study developed a serum Raman spectral classifier using a convolutional neural network. We collected 458 serum spectra (echinococcosis, healthy, cirrhosis, hepatocellular carcinoma) plus 115 test spectra. The CNN achieved 95.7% accuracy. Under simulated Gaussian noise (2 dB SNR), accuracy remained above 80%. Compared to four traditional machine learning models, our CNN showed superior performance with or without noise. This approach enables accurate, robust rapid screening of echinococcosis with clinical potential.