Thermal imaging is a non-invasive and radiation-free technique that captures physiological patterns related to blood flow, inflammation, and metabolic activity. A prior pilot study by researchers demonstrated the feasibility of thermal video analysis for detecting Onchocerca volvulus worm viability using handcrafted features and Random Forest classifiers (AUC = 0.85). This study extends that work by employing deep convolutional neural networks (CNNs) to automatically learn discriminative features from thermal video frames. Six architectures such as VGG16, VGG19, ResNet50, InceptionV3, MobileNetV2, and Xception on thermal videos were used to classify Normal or Abnormal. Results showed that InceptionV3 and VGG16 achieved the best performance with 92% accuracy, supported by balanced precision, recall, and F1-scores. MobileNetV2 and Xception followed closely with 90% accuracy, while VGG19 reached 84% accuracy. In contrast, ResNet50 underperformed at 49% accuracy, indicating poor adaptation to thermal data. Compared with earlier handcrafted approaches, CNNs provided substantial improvements in classification accuracy and robustness. These findings highlight the potential of deep learning for non-invasive, low-cost diagnostic tools in neglected tropical diseases. The study also outlines clinical implications, dataset limitations, and future directions, including temporal modeling of thermal sequences and large-scale validation studies.