Week 5 - Aaron LaViolette

Rats!

A good part of this week was devoted to research. Recently a lab here was looking into a mouse model of how effective a drug (rapamycin) is in treating cystadenomas (solid/cystic tumor lesions) in Tsc2+/- mice kidneys. The thought is that mice with trait will develop an overactive mTOR signaling pathway leading to lesions, and that rapamycin will inhibit this pathway (Palavra et. al., 2007).

The lab had taken T2 weighted images of the mice at two different points in time, some mice being treated with rapamycin and some as a control. If successful, the number and size of the solid/complex lesions should decrease in the rapamycin mice as compared to the control group. Since the lab had done MR images they asked for help of Dr. Prince in evaluation, and since I happen to be advised by him this turned out to be a great project for me for summer immersion.

T2 weighted image of a mouse in a 9.4 T machine.

The most exciting thing about this was that the mice were imaged in a 9.4 Tesla (T) scanner! For some context most, clinical machines for humans are either 1.5 or 3 T (much greater than a typical fridge magnet which is about 5 mT). Since the signal to noise ratio (SNR) in MRI images is proportional to the field strength, these images came out incredibly sharp.

In order to do the analysis, Drs. Prince, Yin and Luo each separately counted things such as lesions and measured tumor sizes for each mouse (blinded to which mouse had what treatment). However, it was interesting to see that despite the high SNR (at least as compared to clinical images) there was still the need for judgment calls to be made about whether certain areas of the kidney were indeed single lesions (as opposed to normal parenchyma or multiple lesions very close to each other). This emphasizes to me the importance of having multiple reviews and doing a reliability analysis.

Since the gold standard for accessing tumor pathology is histology we will have to see how well the MR images match the histology. Should MR and histology match, there are a couple of key advantages. The first is cost. The second is that in order to do histology you must kill the mouse, and so no “before and after” comparisons can be made. MR allows for a “before and after” analysis.

While shadowing Dr. Prince this week at the Columbia Medical Center, I also got a chance to learn about the “liver list.” This list is what is used for the pecking order of liver transplantation. Generally, patients are accessed based on a Model for End-Stage Liver Disease (MELD) score (the higher the score the better you are in the pecking order). However, this is imperfect for certain pathologies such as hepatocellular carcinoma (HCC), and patients can be given points based on the number of tumors and measurements of their sizes (Samuel and Coilly, 2018). With this, there is a “goldilocks zone”: you must have HCC to gets these points, but to many tumors or too large of tumors results in these points not being given. In some sense this is an interesting paradox: you don’t want to have HCC, but if you do you want to fall within the criteria to get the points.

As Dr. Prince pointed out to me a higher score is better for various reasons. The first is immunology (i.e. you have to be similar enough to someone that the transplantation is not rejected). The second has to do with the type of liver you might inherit. Imagine that a drunk, old, motorcyclist happens to crash into a tree one day, and his liver is now up for transplantation (his liver probably isn’t in the best of health). If you are at the top of the list, you’ll pass this liver up and wait for an 18 year old to tragically be hit by a bus (i.e. for a better liver). But if you are lower on the list, you don’t have the luxury of passing this liver up and will have to take the diseased liver.

Most of the measurements are made by MR or CT images, which involve the radiologist measuring the diameter. For various reasons, these measurements are imperfect. So, if a measurement is close to cutoff, the radiologist will make the measurement turn out to be within the criteria for the maximum points instead of outside this criterion. This made me learn that although there is affectively a “look-up table” for points, there is still a human aspect to medicine and radiologists are doing their best to help patients move up the liver list.

This week I also got to see some of the thoughts into the future with radiology. I attended a meeting with Drs. Prince, Shih, Goel and Riyahi, about implementing deep learning to look at patients with autosomal polycystic kidney disease (ADPKD). There are so many things that radiologist is looking for that computers could help with. One aspect is that some of the images can have hidden subtleties to them that are not easily found by the naked eye. For example, in one case I saw this week a patient had a large mass in the liver that was clear on the present study. When compared with the images form a past study, the lesion was indeed visible, but subtle in the image and very easy to miss (unfortunately with this case it was missed). If a computer algorithm could be trained to point out things it “thinks” are suspicious, it may help these subtitles at least be noticed. The other aspect of deep learning is to use it predict outcome. For example, if a patient is imaged they would like to know based on that image what their kidney might look like in 10 years, 20 years, etc. This of course involves lots of data but is an interesting idea. It seems to me that in the future computer algorithms will probably be common place in accessing medical images.

References:

F. Palavra, C. Robalo, and F. Reis, “Recent Advances and Challenges of mTOR Inhibitors Use in the Treatment of Patients with Tuberous Sclerosis Complex,” Oxidative Medicine and Cellular Longevity, vol. 2017, pp. 1–11, 2017.

D. Samuel and A. Coilly, “Management of patients with liver diseases on the waiting list for transplantation: a major impact to the success of liver transplantation,” BMC Medicine, vol. 16, no. 1, 2018.