In order to reduce the clinical misdiagnosis rate of of thyroid disease by using SPECT images, and improve the accuracy of deep learning algorithm in recognizing the features of cross images in nuclear medical image-assisted diagnosis, a thyroid SPECT image diagnosis method based on ResNet model was proposed. Deep Convolution Generative Adversarial Network (DCGAN) and Super-Resolution Generative Adversarial Network (SRGAN) were used to generate images and improve the resolution to make up for the deficiency of training data. At the same time, x_i with the cross-feature image information was added to the residual block output information, and the learning of the cross-feature on the basis of retaining the learned image features, so as to improve the model. As for cross-image features, a cross-training set was proposed to retrain the improved ResNet neural network model that had been trained with a single feature image. The experimental results show that after 100 rounds of iteration, the verification accuracy of the improved residual neural network model trained by the cross-training set is as high as 0963 3, and the verification loss is reduced to 0.118 7, which tends to be stable. The recall rate, precision rate, specificity and F₁ score are all above 93.8% in the recognition results. The improved neural network model and the new training method show higher typical symptom recognition rate for thyroid SPECT images than other methods based on convolutional neural network (CNN), and have reference value for clinical image diagnosis.