Tom

  • Student

Drafts

Machine Learning and Radiology: Friends or Foes?
Tom
Posted on November 13, 2018 at 3:13 pm
Machine learning and artificial intelligence are looming disruptors in the field of radiology. What are leading health systems doing to tackle this issue?

Activity Feed

On November 13, 2018, Tom commented on Open Innovation Driving Growth at Alibaba :

I found this article absolutely fascinating. The most interesting aspect to me was the idea that Alibaba is sponsoring these innovation initiatives without particular end-goals in mind. Many of the articles posted here have outlined how a specific company aims to reach a specific target through various new models (e.g. additive manufacturing in the Navy or crowdsourcing traffic information at Waze). However, given how well-capitalized Alibaba seems to be, their approach is simply to sponsor all types of innovation and then to worry about how specifically to monetize it later in the process. While this may seem like a disorganized approach, it is actually not the first time that a well-capitalized company has used it. In the 1950’s, Ford had an innovation center in downtown Detroit where innovation was simply funded for innovation’s sake, and while some resulting inventions like 4-wheel drive were directly applicable to their line of business, others like the heart-lung machine for cardiac surgery also resulted from this initiative. I’m fascinated to see what results from this initiative at Alibaba, especially in a country as fertile for innovation as China.

On November 13, 2018, Tom commented on How crowdsourcing is changing the Waze we drive :

Great post! As you pointed out, relying on the altruism of the network of editors is a major risk that Waze faces. I’m fascinated with how they’ve “gamified” the editing process in order to create incentives (seemingly out of thin air) in order to encourage people to edit the maps for them. Moreover, these individuals seem to do an accurate job in a timely manner despite receiving no tangible compensation for their efforts. I’m curious whether a more disciplined approach can be applied to machine learning to look at historical traffic patterns, notices of construction on municipal websites, Twitter activity, etc. to automate some of this work and make the platform less reliant on individual volunteers. At the same time, I’m curious what other industries might be able to benefit from creating a game-like crowdsourcing platform in order to recruit the work of volunteers in order to advance their business objectives (and to learn from the human volunteers’ activity in order to create a machine learning-based solution).

This is a great example, and something that a lot of us can relate to! What I found really interesting was how Disney integrates their data collection into the customer experience, making it actually seem really fun for users to give Disney their data. At the center of this is the MagicBand, which visitors actually pay for prior to visiting a park. From a visitor standpoint, using this is really frictionless and, as the name implies, feels like magic. The visitor is incentivized to wear the band and use it throughout their visit, generating high-quality data for Disney at a fraction of the price of other collection mechanisms (especially given the revenue from the bands themselves). There are approaches (such as travel/hospitality companies placing apps on phones) which approximate Disney’s collection mechanism, but none are executed quite as well. Like with many other aspects of the Disney experience, this lesson should be applied to other industries, especially with the proliferation of wearable devices: if executed well, consumers can be highly engaged in the collection of their data, and might even pay you to do so!

This is a fascinating post! I’m curious how this will affect the Navy’s relationship with long-time suppliers to our armed forces. Several large defense contractors are not only an important strategic partner to our armed forces but also have a large impact on the US economy. Maintaining their capabilities is an important part of readiness should the US find itself in a major conflict and have to draw on their skills and capacity to produce new or additional products for defense purposes. These contractors have a symbiotic relationship with our armed forces for the most part, and may hold patents on some of the technologies currently being deployed. Is there a way to deploy additive manufacturing in a way that is sensitive to these existing relationships and their broad impact on our economy? How should the US Navy best partner with its suppliers in order to create a high-quality, cost-effective, and efficient supply chain while also being sensitive to the role of the private defense sector in our economy?

On November 13, 2018, Tom commented on 3D Printing & The Future of Army Resupply in Remote Areas :

This is fascinating and a great use case for this technology! Given the remote location, being able to have a limited stock of supplies for the 3D printer from which a variety of supplies can be produced makes total sense, even if the cost per part is higher (since inventory carrying costs will be low). What struck me as particularly important is the cybersecurity requirements around this as you mentioned at the end of your post. I’m also curious whether any thought has been given to how to best prioritize the parts which would be best made on-site using additive manufacturing vs. being carries as inventory with the other supplies. Understanding this threshold for cost, customization, frequency of use, propensity to break, and other factors may also help create a feedback loop for more robust design of critical supplies for our soldiers.

On November 13, 2018, Tom commented on Using Machine Learning to Optimize Hospital Operations :

This is a fascinating use of machine learning in the healthcare space. There is certainly a need for a better solution than what most hospitals are currently using, and this may very well be it. However, as with most things in healthcare, there is a lot of complexity beneath the surface. I would imagine that data related to several other inputs (e.g. availability, individual variability in efficiency, etc. of cleaning staff required between OR cases) are either currently not collected or are collected in an entirely different system (e.g. a timekeeping system) than the OR case scheduling system. With these factors (e.g. is the person assigned to clean the room generally fast or slow at their job) playing a significant role in turnaround time between cases and general efficiency of the OR’s, there is a lot of work required at each site to turn LeanTaaS into a comprehensive solution. If they can figure out how to do that at a reasonable cost, I imagine that they will quickly build a huge book of business.