Monday, December 19, 2016

Robust to the future of medical imaging

Machine learning can make medical image feature detection, annotation, segmentation, and classification much more robust to changes in the lab and clinic. Traditional deterministic image processing uses fixed pipelines of filters to denoise, simplify, and enhance. The deterministic filter pipelines are designed by experts in image processing who have spent time looking closely at the images, understanding the imaging system that took the images and understanding the goals of the imaging. Also any deterministic feature extraction process will always look at finite number of image features - even if that number is in the hundreds. What happens when looking at a new condition or disease that affects new never before examined features? What happens when the imaging process changed? When a dye is changed? When a MRI pulse sequence is changed? The deterministic pipeline fails in these cases and the pipeline itself has to be changed - meaning rewriting the program to some degree. Machine learning image processing systems only need retraining instead of rewriting making them more robust to future imaging and medical advancements and new areas of research.

Gene editing to see what happens when ...

CRISPR gene editing can increase understanding how diseases happen even before it can be used for treating diseases. CRISPR gene editing is new enough that testing and regulation will take so time before CRISPR gene editing is used in treatments at the doctors office. That won't stop CRISPR from making an impact on medicine soon. Gene editing is already building model organisms of disease - cell lines, lab animals such as mice, that mimic a disease helping researchers understand a disease. Gene editing also lets researchers change a gene to see what happens when something is changed. This transforms genetics and biology from observing genetic variation from the wild to experimenting with genes. Our understanding of the biology will increase rapidly and a deeper understanding of life and disease will let us develop new treatments whether the treatments are gene editing based or not.

Precision medicine even better than personalized

Precision medicine is even better than personalized medicine. We want treatments to work for as many people as possible. Personalized medicine tries to find who a treatment works for and what treatment works for a patient - a worthy goal. Precision medicine seeks to understand why and how a treatment worked or failed. From the why and how the treatment could be improved to work better from a broader, less personalized group of patients.

Data collections over individual studies

Collections of medical data from many studies are better for understanding disease than single studies. Conclusions linking genetic mutations to diseases in individual studies have often been wrong due to small sample sizes. By publicly aggregating the data from many papers together the data becomes larger and more powerful. Merging and comparing cross study data takes powerful normalization tools. Luckily our data normalization tools are getting more and more powerful. One tool is ComBat to remove batch effects in studies: https://www.bu.edu/jlab/wp-assets/ComBat/Abstract.html available from Biocondcutor, https://www.bioconductor.org/ for R, https://cran.r-project.org/.

Cancer prevention

One of the steps we should take in fighting cancer should be to reduce or eliminate government policies that promote cancer. Farm bill subsidies for grains like wheat and corn, and cheap grazing fees for ranchers encourage farmers and ranchers to grow lots of inexpensive high calorie, low nutrition food stocks. These subsidizes in effect take farmers away from growing higher nutrition vegetables. This makes fattening foods cheap and nutritious foods expensive encouraging poor eating habits that increase cancer. We should at least take away government subsidies for cancer increasing food production. Another related step is to stop encouraging driving. National subsidies for the highways and local subsidies for roads makes driving cheap. Local regulations for single use neighborhoods separates our homes from our offices and grocery stores and schools. Walking or biking to work or school or a grocery store or much of anywhere is impossible in almost all of America. Driving a car is the only practical way to leave our homes. I know we don't want it this way because the few neighborhoods where you can walk or ride a bike are very expensive blocking most of us from those neighborhoods. So we drive everywhere, the more we drive, the more we drive cancer through air pollution, through lack of everyday exercise during our lives. Policies at least allowing building mixed use neighborhoods where we can walk and bicycle would drive down cancer. Policies encouraging (or at least not discouraging) nutritious eating habits and exercise in our daily lives could drive done cancer as much as 30% and it's basically a free program because we are already spending the money on things that drive up cancer. Relying solely on high tech medicine and ignoring prevention we can do today in curing cancer will take time and have enormous human and economic costs.