Ari Gleich and Aneesh Zutshi 03/05/2022
Preceptor: Dr. Someshekar Krishna
As part of the Clinical Preceptorships in BME course, we have continued to work with Dr. Krishna, a gastroenterologist, on improving endoscopic practices and technologies. The major task we have been working on so far is finishing our image analysis training. We have each worked with Dr. Krishna on editing real patient cases to identify papillae structures and endothelium that can be used to identify the type of cyst present. This work has been super exciting and it is very interesting to see how a lot of these cases have minutes worth of imaging but struggle to obtain much footage of papillae and epithelium. As we continue to edit patient cases for the AI development study, we hope to think of ways that endoscopic procedures can increase the amount of papilla and epithelium footage obtained.
Outside of image analysis, we have also spent more time asking questions during endoscopic procedures to learn about the technology being used and how it can be improved. One case we recently saw focused on a patient who had a stomach bypass surgery and needed to expand the connection made between the stomach and small intestine. To do this, Dr. Krishna aimed to place a stent there for a few months until tissue formed around the stent mesh to create the connection. However, the standard procedure Dr. Krishna typically employs did not work. Him and his team of 3 nurses had to troubleshoot and try other wires and delivery devices to come up with a combination that would work. Eventually, nurses and a doctor from a neighboring operating room came in and helped with deciding a plan for stent delivery. After 30 minutes, they came up with an approach and were able to implement it for successful stent placement in the patient. It was super surprising for us to see engineering principles applied in a non-traditional way during the procedure. Instead of using engineering technology as it was meant to be used, the clinical team was able to use these devices in unorthodox ways to solve a problem and provide the patient what they needed.Through this procedure we saw the intersection of medicine and engineering. We hope to identify device solutions to the problem Dr. Krishna’s team faced and potentially develop a prototype capable of alternate stent placement.
Ari Gleich Clinical Preceptorship with Somashekar Krishna – Blog Entry 3
Ari Gleich and Aneesh Zutshi 03/26/2022
Preceptor: Dr. Someshekar Krishna
As part of the Clinical Preceptorships in Biomedical Engineering course, we have continued to work with Dr. Krishna, a gastroenterologist, on improving endoscopic practices and technologies. Over the past couple weeks, we finished editing our first set of videos. We went through videos generated via endoscopic ultrasound and edited them so they only included papillary and endothelial structures so we could feed them to the artificial intelligence software. We found this interesting because this process helped us understand what structures Dr. Krishna is looking for while performing imaging. It also was interesting to learn about the process of developing an artificial intelligence algorithm. It must be fed data (training data) and must be fine tuned and able to find patterns in that data before being applied to generalized data sets. Additionally, in some of the videos, we noticed that things would get stuck on the imaging tip and obscure the pancreatic cyst. Therefore, a possible improvement to endoscopic ultrasound would be altering the tip of the probe to make it prevent the adherence of debris. Dr. Krishna mentioned that this tip is high hydrophilic, which I learned in biomaterials should help minimize adhesion. Therefore, other methods to minimize adhesion, like creating a smoother surface should be explored to see if this issue can be minimized. We plan to hopefully do more research in this area to determine if this can in fact be improved.
We also spent much of the past couple weeks conducting research. Dr. Krishna has also expressed to us the challenges of endoscopic ultrasound producing 2D rather than 3D images. During procedures, Dr. Krishna must insert a needle for fine needle aspiration, but this is challenging since depth cannot be determined by a 3D image. After conducting some research, I discovered that tridimensional (3D) endoscopic ultrasound has been created and utilized in certain cases [1]. According to the article, the 2D images can be converted to 3D via linear translation, but this can introduce scans from different planes into the image and lead to reconstruction issues [1]. Additionally, to be able to create a 3D image with the current technology, the surgeon has to rotate the scope around its own axis at the same height via freehand rotation which is extremely challenging to do and introduces artifacts [1]. The only modification the researchers made to existing scopes was embedding a 3D freehand module in the scope [1]. However, the scope is already bulky and this module did not result in clear images. Therefore, we plan to do more research to see if any other techniques exist to go from 2D to 3D and brainstorm ideas of our own.
[1] Saftoiu A, Gheonea DI. Tridimensional (3D) endoscopic ultrasound – a pictorial review. J Gastrointestin Liver Dis. 2009 Dec;18(4):501-5. PMID: 20076829.
Ari Gleich Clinical Preceptorship with Somashekar Krishna – Blog Entry 4
Ari Gleich and Aneesh Zutshi 04/09/2022
Preceptor: Dr. Someshekar Krishna
As part of the Clinical Preceptorships in Biomedical Engineering course, we have continued to work with Dr. Krishna, a gastroenterologist, on improving endoscopic practices and technologies. The past few weeks have been filled with schedule conflicts that resulted in limited clinical experiences. As a result, we have taken time to reflect upon previous endoscopy experiences to begin to improve practices. Many of the issues we saw during procedures revolved around the tip of the probe. One issue was the adhesion of the tip to the calcified cysts, marked by an increased presence of hydroxyapatite crystals. These crystals are known for interaction with the tip through ionic interactions and interestingly enough the highly hydrophilic nature of the probe tip that prevents protein aggregation during procedures may lead to increased interactions with the calcified cysts. As changing the hydrophilicity of the probe tip results in increased adherence of biological substances which impedes imaging, we decided to look into coating the tip with a substance that allows it to retain its hydrophilic nature but limits hydroxyapatite crystal adhesion. Previous work has shown that a modified 14 amino acid peptide is capable of reducing hydroxyapatite crystal adhesion [1]. This decreased adhesion stems from an alanine amino acid modification in the 7th residue which limits electrostatic interactions. A proposed tip coating with the peptide sequence is shown in the image below.
Figure 1: Proposed Tip
Another issue we noticed with the tip was that when the needle is inserted for fine needle aspiration to retrieve a tissue sample, its angle when entering the cyst needs to be at a specific angle. This issue is worsened by the fact that the tip is at an angle compared to the imaging apparatus. This problem has been explored in previous research and an array of different tip to imaging apparatus set-ups are shown below [2]. As shown in Figure 2a, a set-up in which all the directions of imaging and probing are located in the same orientation would be an ideal set-up for physicians that limits the chance of puncture errors. Using this knowledge, we plan to design a probe that combines the tip coating and apparatus arrangement discussed to greatly improve endoscopic practice. Our final design will be done in Solidworks and have material choices/types listed.
Figure 2: Different Endoscope Set-Ups [2]
Lastly, we were able to meet with Dr. Wei-Lun Chao, a professor at Ohio State University who specializes in neural network research. He taught us basic algorithms that can be implemented for machines to distinguish between generic images and also gave us python libraries to get us started on implementing these ourselves. He particularly encouraged us to learn PyTorch to do image segmentation and recognize structures like papillae and epithelium within the pancreatic cysts. We plan to use this knowledge to identify papillary structures. Machine learning attempts to find patterns in data. This is done by providing training data for the machine to learn the neural network. The neural network is then applied to test images to identify papillae and test its accuracy. Once its accuracy has been validated, it can then be applied to normal data. Dr. Chao explained neural networks to us through a lego metaphor that we found fascinating and interesting. With legos, you have all the pieces, but need to be told what to build. For instance, if you have all the lego blocks and are shown an image of a car to build, you can eventually learn how to build that car. A neural network is like that, it gives you all the pieces but it needs to be provided training data to guide it in how to build a series of equations. An example of this is shown through the following PyTorch code (linked below) that was trained with images of the world to recognize things like bikes, humans, and trees in new images [3]. We hope to do something similar, or at least develop the outline for an algorithm to identify papillae.
Figure 3: Example Algorithm Identifying Objects [3]
[1]. https://pubs.acs.org/doi/pdf/10.1021/acs.langmuir.1c02293
[2]. https://europepmc.org/backend/ptpmcrender.fcgi?accid=PMC8871038&blobtype=pdf