#Artificial intelligence (AI) has become one of the most used buzzwords in recent times. AI garnered unprecedented media coverage as well as public attention owing to its unparalleled self-learning capabilities. Convolutional neural networks (CNNs) are building blocks of modern AI, which are revolutionizing several fields of basic as well applied sciences. Health-care sector is not untouched by these advancements, particularly so Dermatology. Several recent studies have shown a glimpse of the future where AI-powered services might revolutionize the diagnosis of dermatological disorders. However, we are yet to see the significant practical impact of AI on public health. Will AI improve the efficiency of over-burdened health-care system, and if yes, to what extent? Will it perhaps replace physicians one day? These are some hotly debated questions of our modern time.
Some of us think that AI will be solution to the most of problems faced by humanity, while others feel AI might be one of the worst problems humanity might ever face!
With starry optimism and a healthy dose of skepticism, this article takes a stock of current developments and ponders over the futuristic outlook of AI-powered diagnostic-therapeutics. Dermatological disorders, in particular, non-melanoma skin cancers are skyrocketing, while the availability of qualified dermatologists is not keeping up with the ever-increasing demand. Consequently, malignancies continue to be a huge socioeconomic burden on the health-care system.
In most of these skin conditions, early diagnosis is known to significantly improve patients’ health outcome. AI-powered services could come to the rescue in at least following two ways: 1. Patient-oriented AI Apps Patient-oriented AI digital systems capable of continuous real-time tracking of the dermatological health could report any aberration to dermatologists. Such a real-time and continuous tracking of dermatological health will not only offer a personalized model of dermatological health but expedite diagnosis of malignancies with great efficiency. 2. Dermatologist-oriented Smart AI assistant Dermatologist-oriented AI systems which can be trained by expert dermatologists to diagnose or help to diagnose skin condi
tions with reasonable accuracy. A well-trained and qualified dermatologist can be served as a ‘benchmark’ to test the accuracy of AI-based diagnosis. In recent times, both these possibilities have been explored by AI researchers to improve the application of CNN in dermatology. Development of patient-oriented AI Apps are still in their infancy; however, some of them are indeed very promising. For instance, SkinVision™ is a Dutch startup which offers an app for Android and iOS. SkinVision™ app claims to perform regular check-ups of skin health and monitor possible malignancies in real-time. The startup received immense media coverage, but it is yet to prove its worth in terms of the diagnostic accuracy. Other notable examples are SkinIO™ and Molescope™. They differ slightly in their market placement compared to SkinVision™: SkinIO™ market themselves to both patients and dermatologists, while MoleScope™ sells a smartphone attachment which aids dermatologist to efficiently diagnose skin cancers. Doctor Hazel is an interesting experimental project operated by a big academic collaboration involving Stanford University and supported by IBM. Doctor Hazel is an AI-powered chatbot, which as of now uses web interface to chat with patients to track patients’ skin health. These patient-oriented AI systems aim to partially or completely replace the dermatologists, albeit in far-fetched future. Replacing dermatologists altogether seems to be unlikely and frankly unsolicited too at the moment. However, it is pretty conceivable that
CNNs will help dermatologists with a preliminary diagnosis. This will not only take away the burden from already overworked dermatologists, but also allow them to focus their expertise on the cases which deserve more attention. Such AI systems will become more pertinent in the future, as these can be made to work on today’s ‘smartphones’. Smartphones brought enormous computational power to almost every household at an affordable price. Given the staggering reach of smartphones (by current estimate approx. 2.6 billion), apps employing local or over the cloud neural networks might prove to be a game changer. Going by current trend, smartphones of future will be lot better in terms of capabilities and affordability. AI-powered apps to track overall skin health in real time and possibly diagnose or help to diagnose malignancies will have a positive and long-lasting impact on dermatological health and socioeconomic impact of dermatological disorders. However, a word of caution needs to be exercised in regard to the authenticity of these AI-based apps. Personally, I found proprietary app marketplaces (particularly Google play store for Android) are littered with numerous apps making tall claims about their diagnostic capabilities and accuracy. A number of AI-based apps catering to health care are going to proliferate quickly and perhaps warrants a regulatory control akin to what FDA would exercise for any diagnostic kits. Certainly, Google
and Apple (two major providers of commercial apps for Android and iOS, respectively) need to invest in curating these apps sans any commercial interests. Although future might bring more of such patient-oriented applications with increasing accuracy and usefulness, as of now, expert-trained AI seems to be a better bet. Indeed, a study published last year in Nature has shown convincingly that CNN is, indeed, as good as dermatologists when it comes to the prediction of skin cancer. A recent study published in Annals of oncology in May 2018 demonstrated that Google’s Inception v4 CNN not only matched the performance of clinical dermatologist, b
ut it outperformed most of the 58 expert dermatologists participated in this particular study. Most of the scientific studies are known to account for their limitations and draw conclusions very judiciously. Pretty much like any other scientific studies, the one in limelight is not free of limitations. Thus, given the limitations of this study, one should not jump to the conclusion that CNN could or would replace dermatologists, at least not yet! Nonetheless, this study holds the promise of future where such CNNs will help dermatologists do their work efficiently. Indeed, authors suggested that the published CNN’s image classification system will offer a valuable assistance to the dermatologists irrespective of their experience (beginners to experts) to improve the accuracy and speed of skin cancer diagnosis. In addition to the academic efforts, big players such as Google, IBM, and Microsoft are throwing their weight behind developing
AI for health care; one can expect advances coming our way with exceptional speed. Continuous improvement through guided or self-learning is the very nature of CNNs and these neural networks are not only here to stay but they are here to improve day by day!
No wonder, this mind-boggling potential of AI sends doomsayer in a frenzy of machine taking over the world, but in my humble opinion, future of humanity is definitely better off with ‘machines being with man’ instead of ‘machine vs. man’!
Android is registered trademark of Google Inc. and iOS is a registered trademark of Cisco, licensed to Apple Inc. Thank you for reading, feel free to give your opinions in the comment section. Follow me @Nandkishor Mule for more such articles.
Comments