And Then There Were Five
Fifteen years after the publication of its first issue, JACR adds a new pillar to its masthead: Data Science.
Recently, a colleague suggested I check out an up-and-coming AI company. Having nothing to lose, the head of my radiology department's imaging post-processing lab and I had a web teleconference with their representatives. They showed us automated segmentation of cardiac cine images, which appeared more robust than our current quasi-automated software and was faster than manual segmentation. Not much faster, but fast enough to keep us listening. The cardiac image quality was good. Maybe too good. I asked, "What about patients whose image quality is suboptimal for any number of reasons, such as arrhythmia-related motion artifact even with arrhythmia rejection? That's where manual post-processing is more challenging and more time-consuming, where we could really use the power of AI." The representatives had to admit that that's where their product fell short, where manual post-processing would still be needed to identify and fix AI segmentation errors.
The next month I went to a CME conference where another vendor was showing a different AI product for a different cardiac imaging application. At my institution, we've been working on finally getting around to offering a certain study for emergency room patients. Our rate-limiting step to date has been post-processing because there are so many competing demands for both urgent and non-urgent studies on our post-processing staff, but resistance to hiring new staff. Given that turnaround time is critical in the emergency department, maybe this could be the answer to not adding more staff while still offering this clinical service.
This company's approach impressed me. Instead of entirely relying on AI software in the cloud, a human reviews and verifies the AI segmentation. "So," I asked, "what's your turnaround time like?" I was informed that in the past year they had cut their time in half. So I pressed for the actual length of time it takes right now. The answer: 5hours. If you notify them about an urgent case, I was told, it could be prioritized, and may only take half that time. The company anticipates ongoing reductions in turnaround time in the future.
Although AI was 0-for-2, it doesn't mean that I won't keep my radar up and tuned for when AI might fit our needs. Therefore, I have started to read more about the topic. However, given everything on my plate right now, I don't have too much time to devote to it, so I'd prefer one-stop-shopping for my AI-in-radiology needs. Fortunately, it's as if folks at ACR and JACR have read my mind.
ACR is starting a new Data Science Institute. The DSI even has its own blog. JACR initially started with three pillars (Health Services Policy and Research, Clinical Practice Management, and Training and Education), subsequently added a fourth (Leadership), and now, if you go online or look at the cover of the latest issue, has added "Data Science". You can read Editor-in-Chief Bruce J. Hillman, MD's rationale for this timely addition in the March 2018 issue. If you want to dive deeper, explore the JACR special issue on data science that accompanied the March 2018 regular issue. This introductory article in the special issue by editors Safwan Halabi, MD, Charles Kahn, MD, and JACR Deputy Editor Ruth Carlos, MD, is a great springboard to get started.
In the meantime, while I wait for AI to mature, I'll be reading everything coming out of the Data Science Institute, and when the time is right for us, we'll make the leap. But for now, we are hiring more post-processing staff.