Healthcare — the hottest area in AI development
According to CB Insights, AI startups in healthcare have raised more capital than all other industries. AI healthcare startups raised $4.3B across 576 deals since 2013 — and this pace of funding continues to grow, as the first two quarters this year consecutively saw the highest investment levels in the sector to date.
Furthermore, AI-powered applications, created both by leading companies in the healthcare space as well as by this new generation of startups, are not just staying in the realm of R&D or experimentation. These applications are beginning to achieve government approval and to be used in doctors offices, hospitals and homes around the world.
Fast-Track Government Approval
In America, the FDA has emerged as a tentative supporter of AI applications in medical imaging and diagnostics, given recent exciting advances in the field.
The FDA began a pilot program in January of this year, under which healthcare software developers could be ‘Precertified’. The logic here was that software products can be quickly adapted to respond to glitches or other safety concerns, so if a company has demonstrated that they have produced dependable and reliable software-based diagnostic tools, they can be pre-certified to allow for a faster pace of development. The FDA now hope to expand this program to include AI software, given its great potential.
This fast-track regulatory approval could lead to major opportunity for the over 70 AI startups in the healthcare space that have raised capital in the past 5 years.
Even before this pre-certification program is in place, the FDA has been approving various AI software applications. In April, the FDA approved an AI system that is able to diagnose patients with diabetic retinopathy, a research area that Neuromation data scientists have previously published on. This software was given “breakthrough device designation” to speed its approval and allow it to reach the market more quickly.
Other AI applications to recently receive approval include one by Viz.ai for analysis of CT scans to identify potential stroke risk in patients. Startup Arterys received FDA approval for a cloud-based AI platform for analyzing cardiac medical images, and this year received approval for software that automates detection of liver and lung lesions in cancer patients.
Another exciting development area involves consumer-side AI-enabled image analysis via smartphones. the FDA recently approved a home urine analysis kit from Dip.io, which uses a smartphone camera to analyze a dipstick. Another startup SkinVision uses the smartphone to analyze skin health at home.
The pipeline of new AI-enabled diagnostic applications is extremely deep, so if anything we should expect to see an acceleration in the number of applications hitting the market in the coming years.
AI has a demonstrated an exceptional ability to detect patterns in data that may have been previously unrecognizable, or even if they were known, were too difficult to detect effectively by medical practitioners.
Several startups are using AI to develop blood tests for detecting various cancers. AI is also now being used for the analysis of secondary risk factors for diseases, which were previously considered too subtle to use as an accurate means of diagnosis. These include AI applications for detecting heart problems, age and smoking history by analyzing patterns in the retina; or diagnosing coronary artery disease via analysis of a patients voice.
Pharmaceutical industry leaders Amgen, GlaxoSmithKline, Merck, Novartis, Pfizer and Sanofi are all pursuing partnerships with AI startups in the area of automated drug discovery.
Techniques being employed in these partnerships include modelling of diseased and healthy human cells to conduct analysis of the impacts of various potential drugs at a scale previously impossible. Another approach is to use billions of pieces of unstructured and siloed data from research papers, patents, clinical trials and patient records to identify previously unnoticed patterns that may lead to new drug candidates. Others, including Neuromation, are using generative adversarial networks (GAN’s) to create entirely new molecules and to ascertain their potential activity. These techniques could lead to faster, cheaper and more effective drug discovery
Neuromation Activity in Healthcare
Neuromation is actively involved in Healthcare research, having published 5 scientific reports in reputable journals on the application of artificial intelligence in the fields of medical imaging and drug discovery. These papers include studies on the use of deep learning in identification of diabetic retinopathy (a leading cause of blindness), breast cancer image analysis and detection of angiodysplasia (a common cause of gastrointestinal bleeding and anemia).
Neuromation also works together with healthcare industry participants to develop image recognition and facial recognition applications for early disease detection and diagnosis. Neuromation is a leader in the creation of synthetic data for accurate identification of rare and hard-to-catalog diseases for which data sets may not exist.
In the area of Pharmaceutical research and drug discovery, this year Neuromation published a paper on 3D Molecular Representations in the journal Molecular Parmaceutics in conjunction with partner Insilico Medicine.
Finally, Neuromation provides services in the area of creation and training of machine learning models to generate and classify molecules most likely to have desired properties for molecular biology and drug discovery applications.
Neuromation Published Papers on Healthcare (medical imaging and drug discovery):
- Diabetic Retinopathy detection through integration of Deep Learning classification framework, A. Rakhlin, bioRxiv, June, 2018, DOI: 10.1101/225508, https://www.biorxiv.org/content/early/2018/06/19/225508
- Deep Convolutional Neural Networks for Breast Cancer Histology Image Analysis, A. Rakhlin, A. Shvets, V. Iglovikov, A.A. Kalinin, Proceedings of the 15th International Conference on Image Analysis and Recognition (ICIAR 2018), DOI: 10.1101/259911, https://www.biorxiv.org/content/early/2018/04/02/259911
- 3D Molecular Representations Based on the Wave Transform for Convolutional Neural Networks, D. Kuzminykh, D. Polykovskiy, A. Kadurin, A. Zhebrak, I. Baskov, S.I. Nikolenko, R. Shayakhmetov, A. Zhavoronkov, Molecular Pharmaceutics, 2018, DOI: 10.1021/acs.molpharmaceut.7b01134, https://pubs.acs.org/doi/10.1021/acs.molpharmaceut.7b01134
- Angiodysplasia Detection and Localization Using Deep Convolutional Neural Networks, A. Shvets, V. Iglovikov, A. Rakhlin, A.A. Kalinin, accepted to IEEE ICMLA 2018, bioRxiv, DOI: 10.1101/306159, https://www.biorxiv.org/content/early/2018/04/23/306159
- Automatic Instrument Segmentation in Robot-Assisted Surgery Using Deep Learning, A. Shvets, A. Rakhlin, A.A. Kalinin, V. Iglovikov, accepted to IEEE ICMLA 2018, bioRxiv, DOI: 10.1101/275867, https://www.biorxiv.org/content/early/2018/03/03/275867
- Pediatric Bone Age Assessment Using Deep Convolutional Neural Networks, V. Iglovikov, A. Rakhlin, A.A. Kalinin, A. Shvets, bioRxiv, DOI: 10.1101/234120, https://www.biorxiv.org/content/early/2018/06/20/234120