AI4All Day 6: Image Processing, AI in Radiology, Career Networking

Nidhi Parthasarathy
5 min readAug 20, 2022

--

Nidhi Parthasarathy, Tuesday, July 5th, 2022

All About Images…

In today’s session we talked about images. The lecture covered a variety of topics including image representation, image manipulation, features, feature extraction, and pooling. We started the lecture by talking about how images are represented. Ayush showed us a really nice diagram of a Matrix of Matrices and explained how pictures are represented with 3 colors, red, green, and blue and how each one of these colors are represented by a matrix with intensity values between 0 and 1.

The intensity values of an image in RGB [source]

He then explained how these matrices together create a picture and discussed alternate, but less-frequently used, color models like the Hue Saturation Value Model (where all colors/hues can be represented on one axis). After seeing images are represented, we talked about image manipulation (cropping, tinting, concatenating, rotating, etc.). Next we talked about features and how to extract them. We learned about what features we should use based on shapes, patterns, and colors, and how we could use filters to find patterns. Finally, we learned about max pooling and average pooling to make images smaller and more pixelated. We also talked about shallow and deep features and the difference between ML and DL.

In the second portion of the cohort groups, we did more hands-on work with numpy and matplotlib and continued the google collab from last week.

Radiology and AI (Medical AI)

In the afternoon session, we heard a talk from Greg Zaharchuk, a Professor and radiologist at Stanford who works a lot with Deep learning and Imaging. He spoke about AI transforming medical engineering. He began his lecture by explaining what radiology was: a speciality of medicine that is learned after medical school during residency that focuses on obtaining and interpreting images of the inside of the body to obtain critical information on how to identify disease and treat patients. He also discussed MRI (magnetic resonance imaging), a technique that is very useful, but also very expensive, and needs a lot of electricity and cooling, and human labor.

He then talked about the radiology value chain and how imaging is the leading healthcare advance, but also very costly and inefficient. He then discussed the use of AI in radiology including heat maps and saliency maps and how they could be useful. He also brought up some of their deficiencies, for example AI models can take short-cuts or identify bad patterns that are not completely true (for example, a group of pictures where the AI model just used the hospital tag to identify diseased people).

He also talked about how to make MRIs faster — super resolution and image transformation to reduce patient risk. He discussed MR contrast (using a material called Gadolinium) and some of the side effects for some patients (e.g, kidney failure) and some of the challenges and opportunities here. He also discussed PST — how deep learning can be used to lower doses and enhance imaging with deep learning — as well as other work he has done on reducing strokes (with Yannan Yu).

The Importance of Networking and Where to Start

The last session featured a talk on networking from Yuliya Mykhalovka, from Greylock Talent Team. She discussed her excitement about startups and product-market fits and business needs in startups and when people should choose startups based on their interests. She then talked about practical networking. She shared nice facts and anecdotes: 70% of people find their jobs at LinkedIn. It is not what you know or who you know, but who knows what you know. She mentioned that it would be good to have something clear to show including your hopes and dreams so your network could help you achieve those.

She then shared a treasure-trove of ideas: Networking is relationship building. It is important to be successful and happy where you are, so how do you start?

The first step to starting is to identify your goal — having specific shareable goals that allow your network to help you is important. When you seek out a mentor, it is important to understand what kind of support you are looking for, and build trust.

You need to have a sense of mutuality so you can identify places where you can help them as well. Introductions matter. Make sure you share your goal and mention your credibility. You need to do your work. Provide a blurb about yourself and your resume, and do your research.

She also talked about how networking should not be transactional. You should stay in touch and stay on their radar and give updates on what you are up to so they can think of you when they have opportunities.

If it is an in-person event, who do you approach? Approach friendly nice people who look like they are outgoing, or talk to the new kid who is also looking for other people to talk to. Or approach groups where they have a conversation going and you can try to contribute. Some starting questions: what got you here? What is your connection to the program? What is your favorite food? She ended by emphasizing how once you get help from your network, you should still follow-up — thank you notes and stay-in-touch notes so they are still looking out for you.

For job searching, do not apply to a bunch of jobs online. 80% of time should be spent networking and only 20% of time on online jobs and applications. When you send your resume, the top third of your resume should have the most important things. When people skim through it, they usually lose interest towards the end, so it is important to get to the most important aspects first.

What an useful session! And an overall very nice day!

Read on for Day 7.

--

--

Nidhi Parthasarathy

Highschooler from San Jose, CA passionate about STEM/technology and applications for societal good. Avid coder and dancer.