Speaker 1: Cloud computing has helped many industries innovate to new heights, and health care is no exception. In previous videos in this series, we looked at how the Cloud Health Care API can help you store and access health care data in Google Cloud. In this episode, we'll explore how the Cloud Health Care API can be used to store, retrieve, and analyze medical imaging data. There are several key challenges that health care professionals face when dealing with medical imaging. First, you need to ensure HIPAA compliance in clinical workflows to ensure patient privacy. Second, researchers often have to learn about new technologies, which can be confusing and expensive. And third, it can be difficult to leverage this data to gain important insights using big data and machine learning. Luckily, the Cloud Health Care API addresses these challenges with one of its endpoints and comes packed with other helpful features for medical imaging analysis. It supports digital imaging and communications in medicine, also known as DICOM, an international standard file format used for storing and transmitting medical images across technologies. This could include x-rays, MRIs, ultrasounds, and more. It can also help you save money by enhancing or even removing the need for specific on-premise software that requires expensive licensing fees. And it makes it easy to scale customer architectures while maintaining low latency and high performance. The Cloud Health Care API also helps you leverage the power of machine learning by integrating well with Vertex AI, Google Cloud's unified AI platform. And finally, it connects easily with open source tools like the Open Health Imaging Foundation Viewer, also known as the OHIP Viewer, which lets you view medical images for the purpose of analysis. This is because the Cloud Health Care API exposes the DICOM store through a DICOM web interface. Users who might be interested in using the Cloud Health Care API for imaging include radiologists who may want to view images, researchers and data scientists who may want to use images for diagnostics, and IT decision makers and clinical organizations who are looking to reduce costs and improve storage scale and elasticity. Let's take a look at an example of how the Cloud Health Care API can be used to build a spine detection machine learning model using a small set of DICOM CT images. First, images are ingested into a DICOM store. A data store is simply a place to store a certain type of health care data, so a DICOM store is a place to store DICOM medical images. Next, we can view the images from the DICOM store using OHIP, an open source medical imaging and viewing tool that integrates directly with the Google Cloud Health Care API. Images can then be parsed into metadata and streamed to BigQuery for further analysis. BigQuery is Google Cloud's large scale data warehouse that's great for storing, analyzing, and visualizing large data sets. With metadata ingested into BigQuery, it becomes much easier to search across a large amount of image metadata that wouldn't be readily searchable in other systems. For example, we could search for the most recent 20 images of lung cancer diagnosis. Once our BigQuery search is done, we can use the corresponding DICOM web path to find the specific image for further analysis. The next step is to use filtered export to export specific image instances to Cloud Storage, which is used to store file objects in the cloud. Filtered export is useful because you may want to export specific images from a larger data set to Cloud Storage and convert them from DICOM to PNG or JPEG for further analysis. Once the images are in Cloud Storage, we can then start training our machine learning model using these images as our test data set. First, we can import the images into Vertex AI as an object detection data set. Vertex AI is Google Cloud's unified machine learning platform that makes it easy to build and train machine learning models on Google Cloud. Then we can label these test images directly in Vertex AI using the Cloud Console. Here's where we can label images that have a spine in them. Once our test data set is ready, we can start to train our prediction model. We can use AutoML, which actually does most of the work for us. All we have to do is provide a label training data set, and Google Cloud automatically builds us a machine learning model that leverages its powerful computing resources. No prior knowledge of machine learning is required. Once AutoML finishes building the ML model, we'll get an online prediction endpoint that can be used to predict whether or not new images have spines in them. The final step is to use the OHIP viewer to view images from the Healthcare API DICOM store and use the online prediction endpoint to give us our ML predictions. Here we have our image being shown in the OHIP viewer. Under Predictions, we can click Find Spine to call the online prediction endpoint hosted on Vertex AI. And since a spine is detected, it is outlined with a box directly in the image viewer itself. We can also take a look at the JavaScript console to see what's going on behind the scenes when we hit Find Spine. There are just two API calls. The first API call is a render request to the Cloud Healthcare API, which we use to get a rendered view of the image. The second API call is to an online prediction endpoint hosted on Vertex AI. This is what we use to predict whether or not there's a spine in the image. The response we get back includes confidence scores as well as bounding box information for the spine detected. In the end, OHIP renders the image as well as the bounding box on top of the image. As you can see, the Cloud Healthcare API provides a powerful platform to help you analyze medical imaging data. An important feature is that it integrates well with Google Cloud products like BigQuery, Cloud Storage, and Vertex AI, giving you new ways to gain invaluable insights about medical imaging data while complying with HIPAA and other government regulations. To learn more, visit cloud.google.com slash healthcare. To get started, you'll need to have a Google Cloud project. If you don't have one, we've included a link to a trial account with free credits in this video's description, along with other helpful resources. And friends, if you found this episode helpful, please subscribe to the channel to get notifications of more health care episodes. Cheers.