Introduction: Accurately understanding the stages of cancer progression has a great impact on choosing the treatment method and the possibility of the patient's recovery, and predicting the time of metastasis is one of the most important factors in choosing the treatment method. In this research, using the periodic tests database of more than 5000 patients with colon and liver cancer, we have investigated the relationships between the results of these tests by using the machine learning method and optimizing the learning algorithm by using complex systems.
In this method, we used image processing and linear regression to analyze the results of the patient's medical images and statistically analyzed the results of biochemical tests. The result of this work was the correct prediction of the start of metastasis with a rate of 53% among 50 test samples. It should be noted that the result of our prediction estimated a period of 38 days
Methods: Our method for the analysis of medical images (MRI) was to use image processing using regression and combining supervised learning methods and modifying the neural network using the equations obtained in the data analysis by unsupervised learning method. Also, the results of biochemistry experiments were done by statistical analysis method using unsupervised learning and network optimization using Lotka-Volterra equation.
Results: By using a neural network trained to check the information of periodical tests and MRI images of 50 patients whose test results and the date of the beginning of metastasis were known in the patient records, we successfully predicted the time of the beginning of metastasis with 54% success. Then with optimal Creating the algorithm and using Lotka-Volterra equation, we re-examined these patients, and the prediction results were successful up to 76%.
Conclusion: The use of statistical data, image processing and the combination of machine learning methods along with optimization using Lotka-Volterra equation can help predict the metastasis time of new patients from the records of previous patients. As the number of data increases, our artificial neural network can improve and train itself with higher accuracy and help predict the time of metastasis with much higher accuracy. In this case, the doctor can choose a more appropriate treatment method using the resulting information. Also, our investigations show that by using this artificial neural network, it is possible to predict the possibility of cancer in susceptible patients.
Keywords: Artificial neural network, cancer, predicting the start of metastasis, complex systems