Objective To propose a high-precision deep learning-based image assessment method  of osteosarcoma chemotherapy efficacy for clinical treatment, as existing methos have low accuracy of osteosarcoma assessment.   Methods The low incidence of osteosarcoma led to the small scale of its imaging data and the problem of imbalance in data categories. This study combined deep learning with clinical medical information, combined the bone sarcoma generation module of BoneGAN and the scale lesion information capture module, and proposed OMLA-Net, a deep learning assessment network for chemotherapy effect of bone sarcoma based on multi-scale lesion attention network, which achieved computer-aided bone tumor assessment with integrated data augmentation and focused lesion information through pre-training and generalized loss training. Results  In this study, 40 cases of osteosarcoma MRI data were used as the basis for the comparison test on the generated dataset, and the OMLA-Net assessment outperformed the SOTA method  Conv-LSTM-GAN in terms of the assessment effects such as accuracy and F1 scores, and the difference was statistically significant (P<0.05); the subsequent K-fold cross-validation ablation experiments further demonstrated the effectiveness of each module proposed by OMLA-Net.   Conclusion   OMLA-Net can effectively perform the impact assessment of chemotherapy effect on osteosarcoma, which provides a new idea for subsequent clinical application.