Analysis And Evaluation Of Corneal And Bone Tumor Images Based On Machine Learning

As an auxiliary diagnostic tool,medical imaging plays an important role in doctors' diagnosis and treatment.With the development of information technology,the amount of medical image data has increased significantly.Based on intelligent information processing technologies such as machine learning,the digital features in medical images are extracted,and their value is tapped to assist doctors in making more accurate diagnosis and evaluation.In order to further analyze and utilize the characteristics of medical images,this paper respectively proposes two analysis and evaluation algorithms based on machine learning algorithms for two types of medical images with large data volume and small data volume.The corneal images with large data volume and bone tumor images with small data volume are selected.In this study,an analysis and evaluation algorithm based on clustering algorithm is proposed for corneal images with large data volume,and an analysis and diagnosis algorithm that uses generative adversarial network to augment the training set is proposed for bone tumor images with small data volume.The corneal topography in corneal images can fully reflect the refractive distribution of the anterior surface of the patient's cornea.The current method to evaluate the treatment effect of Orthokeratology has rarely used the data features in the corneal topography,and is relatively subjective.In order to improve the shortcomings of the existing evaluation methods,this paper analyzes the corneal topographic map based on clustering algorithm,and proposes an objective and quantifiable evaluation algorithm of Orthokeratology lens fitting.The core of our evaluation algorithm uses clustering to obtain the precise set of edge points of the treatment zone,and then find the position of the treatment zone by the least square algorithm to evaluate the patient's fitting after wearing the lens.The experimental results show that,compared with the existing Orthokeratology lens fitting evaluation method,the proposed algorithm can more objectively quantify the decentration magnitude of the patient after wearing the lens,and has good consistency with the evaluation results of professional ophthalmologists.Bone tumor images contain many data features,but due to the small amount of bone tumor images,they cannot meet the data volume requirements of general machine learning algorithms.Therefore,this paper proposes an evaluation and diagnosis algorithm based on bone tumor images,which contains a generation part and a diagnosis part.The generation part uses the generation capacity of the generative adversarial network to increase the amount of bone tumor images,and proposes a conditional generative adversarial network model A-CSGAN(Attention Based Content and Style Generative Adversarial Networks).This model introduces attention mechanism in the discriminator to improve the discriminatory ability.It also uses the content loss and style loss to improve the clarity and diversity of the bone tumor image generated by the model.In A-CSGAN,conditional input increases supervision information to the network.The diagnostic part uses a deep convolutional neural network to build a diagnostic classification model,and the training set composed of generated bone tumor images and real bone images improves the performance of the diagnostic model.