What is the meaning of life? Are robots going to take over humanity? I’m sure I am not the only one kept up at night by such obscure thoughts. We can’t possibly answer the first question in due time, that would only create unnecessary head scratchers, but for the second one I can only say this: probably not…yet? A paper published in the American Journal of Surgical Pathology describes the impact of deep learning assistance on the histopathological review of lymph nodes for metastatic breast cancer, which further sheds some light into how exactly artificial intelligence (AI) is going to either help or replace human labour.
The charm of creating intelligence preoccupied humankind all the way back to the time when Talos protected Europa from invaders. Nowadays, state-of-the-art computing resources allow Google to make progress in a branch of AI called deep learning. This software attempts to mimic the activity in layers of neurons in the cortex. It basically learns to recognize patterns in digital representations of sounds, images, and other data. Moreover, it is proving to be of great help in the medical field, already surpassing specialists in speed and accuracy. In pathology, reviewing lymph nodes for breast cancer metastases can be a truly tiring job; that’s where the Lymph Node Assistant (LYNA) comes especially in handy, considering the difficulty and, most of all, the importance of detecting early micrometastases.
The study, a bout of strength between humans and machines: 6 pathologists completed a simulated diagnostic task in which they reviewed lymph nodes both with and without the assistance of LYNA. The use of LYNA made the task easier and halved average slide review time. Moreover, algorithm assistance significantly increased the detection of micrometastases (from 83% to 91%).
It is important to note that algorithm assisted pathologists actually proved higher accuracy than both the algorithm and the pathologist alone. Why is that? The perks of AI have already been recognized: some can even exceed a pathologist’s sensitivity for detecting individual cancer foci in digital images. However, this gain in sensitivity comes at the cost of increased false positives. Also, it is limited to the task for which they have been specifically trained. On the other hand, the pathologist’s experience and understanding of the clinical context proves invaluable for discerning false results. Hence, combining the sensitivity of the software with the specificity of humans actually brings the best of both worlds to accurate results.
We have barely begun to understand the strengths and weaknesses of these algorithms, so the potential clinical utility has not yet been thoroughly examined; still, the future looks bright! Even though shows like “Black Mirror” paint a gruesome outlook for the distant future, one cannot deny the beauty and the necessity of technology in our lives, especially since we are lucky enough to be witnessing fiction slowly turning into reality. But next time you want to jokingly hit your Roomba or toy with your Google Home, you may want to think again…they might remember.
Follow this link to read more on this particular subject: https://journals.lww.com/ajsp/fulltext/2018/12000/Impact_of_Deep_Learning_Assistance_on_the.7.aspx
Silvia Dumitriu – SOMS Wisp of Sciene