AI predicts survival
American scientist Hugo Aerts and his colleagues developed a deep learning model that used tomographic images of tumors of patients with non-small cell lung cancer in the dynamics of treatment results.
Lung cancer is the leading cause of death from cancer worldwide. Non-small cell lung cancer accounts for about 85 percent of all lung cancers. The standard assessment of diagnosis and response to therapy for patients largely depends on measuring the maximum diameter of the tumor, which is subject to change over time.
To understand whether it is possible to make more accurate predictions about the results of therapy, scientists have created a deep learning model. They transferred data from the ImageNet neural network created by researchers from Princeton and Stanford Universities and trained their models on a series of CT scans of 179 patients with stage 3 non-small cell lung cancer after chemoradiotherapy. The images showed patients before treatment, one, three and six months after treatment. A total of 581 images were obtained.
The researchers analyzed the model's ability to make reliable predictions of the outcome of the disease using two data sets: a training set of 581 images and an independent set of 178 images from 89 patients with non-small cell lung cancer after chemotherapy and surgery. The performance of the models improved with the addition of each subsequent set of images.
The accuracy of the forecast increased to 0.74 after adding all available CT scans. Patients assigned by the model to the low-risk mortality group had a six-fold increase in overall survival, in contrast to patients assigned to the high-risk group.
In comparison with the clinical model, which uses parameters such as the stage of the disease, gender, age, tumor size, the presence of bad habits, the deep learning model has shown great effectiveness in predicting distant metastasis, progression and local recurrence.
The only drawback of this research method is a small data set that needs to be increased in order to evaluate it in further clinical trials.
The proposed deep learning model will further help clinicians adapt treatment plans for individual patients and help collect statistics in various risk groups for clinical trials.
Article by Xu et al. Deep Learning Predicts Lung Cancer Treatment Response from Serial Medical Imaging is published in the journal Clinical Cancer Research.
Elena Panasyuk, portal "Eternal youth" http://vechnayamolodost.ru / based on AACR materials: A Deep-learning Model May Help Predict Lung Cancer Survival and Outcomes