University of Alabama at Birmingham researchers have developed a machine-learning framework that could improve the effectiveness of cancer screening tests. The team led by Jianfu Lio ran a clinical trial that examined current breast cancer screening test values versus a machine-learning algorithm that had been trained on the treatment regimen. Results of the study led to the machine-learning model demonstrating promising predictive power for detecting cancer cells in the bloodstream. The result of the study was published in PLOS Computational Biology.

Recently medical school graduates who were interested in using machine learning to improve cancer screening tests presented with the problem of determining the best method for detecting cancer cells in the blood. Many candidate tests are based on similar algorithms for detecting cancer cells in the blood. However the interpreting the results of these tests has proved to be difficult. One possible way for the testing community to improve their ability to detect cancer is to include more info about treatments and possible side effects of cancer in the algorithm. For example a test that looks for cancer cells in the blood might include information on all cancer-causing proteins in breast cancer cells compared to the ten proteins detected using the human test. Such a feature would increase the ability of both algorithms to detect cancer cells in the blood and help to improve the effectiveness of cancer screening tests. Hopefully such an algorithm will help increase screening test accuracy.

Jingfu Lio assistant professor in the Department of Biomedical Engineering and Mechanics in the Jacobs School of Engineering.

Analysis of current screening results using both algorithms proved to be helpful for the researchers. From the blood of 94 participants 39 cancer cells were identified. The researchers then classified the detected cells into five groups: detection group routine group detection progression group and detection supertest group. The researchers were able to determine that the algorithm predictions were superior to that of the human based on the classification which demonstrated improvements over the algorithm performed on human.

Jingfu Jenny Liu a postdoctoral fellow in the Lab of Artificial Intelligence in the Department of Bioengineering in the Jacobs School of Engineering and Applied Science.