The hashtags of Ai4 Healthcare 2019 Conference at NYC are #ai4 #ai4healthcare. [Image Source: Engy Fouda]

Ai4 Healthcare 2019 proves that AI can save lives

Engy Fouda
6 min readJan 10, 2020


The conference speakers demonstrated how AI can improve healthcare with use-cases. However, many clinicians look to it with wary eyes.

NEW YORK CITY — Clinicians and engineers argued about the effectiveness of using AI in the health industry through technical and business tracks for a successful, packed, two-day conference, “Ai4 Healthcare,” which started on Nov. 11, 2019.

Four presentations were running in parallel to accommodate the 86 speakers from various sectors of the healthcare industry. The presentation halls were packed. During Phillips and OpenPharma-Opioid presentations, the rooms were full of plenty of attendees standing.

“We had about 400 attendees at Ai4 Healthcare,” said Jessica Gallagher, public relations lead at Ai4.

They presented their machine learning models to help physicians, radiologists, and pathologists. Some of the goals were to diagnose cancer, predict the cause of death due to opioids, decrease suicide, predicting hepatitis C infection, and forecast lung health.

The conference venue included live demos to the sponsors’ products and startup showcases. About 34 companies sponsored the conference, including Microsoft, .dotData, and AWS.

Breast Cancer

“I’m a clinician, and I really share your excitement about the ability to predict early…In the long term, it [AI] doesn’t save lives. So, as you perfect these models, should there be a broader plan to see if it’s actually making a difference?” said one of the attendees to Krzysztof Geras, assistant professor at NYU school of medicine and NYU Center for Data Science during his lecture about “Deep neural networks improve radiologists’ performance in breast cancer screening.”

Geras agreed with the attendee that now the model is still not as accurate as radiologists. However, it is precisely comparable to about 77%.

A comparison between the model developed at NYU is about 77% precise to human radiologists’ performance. [Imafe Source: Engy Fouda]

He presented many challenges that his team is facing to optimize machine learning models to diagnose breast cancer correctly, prominently, the lack of public data, and positive examples other than the constraints of the academic research versus companies.

“Definitely, in the longer-term, I expect that this field is going to drift more into doing things like prognosis as well rather than diagnosis.”

Mortality Data Insights On Opioids

Although clinicians argued the speakers at various lectures about the feasibility of AI models in prognosis and diagnosis, the pathologists and the engineers seemed to agree on everything concerning death, which was evident during the “Mortality data insights on opioids and substance abuse” lecture.

Dr. Brandi C. McCleskey, forensic pathologist, Jefferson County Coroner’ s medical examiner’s office; assistant professor, department of pathology, University of Alabama in Birmingham presented the implementation and real results from a national mortality reporting application, called Mortality Data Explorer — built on the OpenPharma platform relying mainly on details listed in death certificates.

McClesky said this platform could produce which two or three substances co-worked together and caused death by using natural language processing (NLP). She said she could run quickly and show you how many times Tramadol is linked to death. “My clinical colleagues think Tramadol is a lower risk opioid.”

McKlesky showed a word cloud be generated based upon the frequency of occurring a particular cause of death.

The most conspicuous words are coronary artery atherosclerosis, hypertensive heart disease, followed by heroin toxicity, then gunshot wound of the head and natural causes, then Fentanyl toxicity.

She said death certificates do not certify drug death as a suicide. “Alabama is one of the worst states for prescribing opioids.”

She said the platform could save future cases by predicting if a teenager would overdose for homicide if linked to other clinical organizations.

“I know when a dirty drug hits the streets,” McClesky said.

Real-Time Insights with 2D and 3D Imaging

At Hub2 hall, there was a contest-like between the speakers to prove that each one’s model is better in predicting cancer than the others.

Georgios Ouzounis, Vice President of Data Science at ElectrifAi, stood solid through this contest. He presented a complete framework that uses segmentation, ANN, and DICOM to answer visually to queries with 2D and 3D models of the organ in real-time.

Georgios Ouzounis, Vice President of Data Science at ElectrifAi, presented a complete framework that uses segmentation, ANN, and DICOM to answer visually to queries with 2D and 3D models of the organ in real-time. [Image Source: Engy Fouda]

“We do not use Tensorflow; we use different SAS products; one of them is Health AI,” he said.

Ouzounis presented plenty of examples; some of them were liver tumor ablation, the brain — aneurysm, and hip joint, and renal calculi.

Sample outputs of ElectrAi’s system using 2D and 3D modeling. [Image Source: Engy Fouda]

However, he said, “These are still empirical results as no doctors verified it yet. We are working on that now.”

During the presentation, Ouzounis listed the same challenges that there is a shortage in the public data to train the models and convincing the clinicians to rely on AI.

“Doctors do not trust it,” Ouzounis said.

He said their framework is not only for healthcare but for any businesses that need real-time insights to maximize profit and decrease risk.

Startups — Asthma

Other than the presentations, there was a live demo to Microsoft’s new augmented reality glasses to explore the human body to help in remote surgeries.

Moreover, Vitalflo was one of the startups showing its product to use analyzing the air surrounding asthma patients to avoid a new attack. Their product consists of two components an inhaler and an air analyzing box.

Vitalflo’s products: the inhaler that analyzes Asthma patient’s breath and a box to analyze the surrounding air to predict if there is a close attack. [Image Source: Engy Fouda]

The patient breaths in the inhaler, which analyzes the breath, and sends the analysis with Bluetooth and wi-fi to Vitalflo software.

Along with the air analysis from the box, the software will alert the patient if there is a probability of having an attack. Then the company’s care-team will call the patient to direct him on how to avoid it, for example, by changing air filters or getting out of the current location instantly.

“We currently serve 50 patients but will reach 2000 patients by 2020,” said Ryan Kelley, Vitalflo Head of Product.

Kelley said the company is trying to have the medical insurance to cover the equipment and the care in the United States and look forward to expanding to other countries.

Startups — Mental Health and Suicidal Thoughts

Another impressive startup showcase was Supportiv, that uses NLP in mental healthcare.

Pouria Mojabi, Supportiv Co-Founder, said, “We make AI matching between patients and support groups according to struggles which moderated by a human monitor.”

Mojabi said the groups’ monitors are trained to deal with people with suicidal thoughts and communicate instantly with therapists collaborating with Supportiv.

“We do not do any labeling,” said Mojabi.

The company moderated 2M conversations in 2018, he said.

In conclusion, engineers’ main challenges are the shortage of training data and the clinicians’ adoption of AI.

During “Ai4 Healthcare,” clinicians who used AI were encouraging others to try it.

“AI will become clinically relevant when (and only when) it is embraced by clinicians,” said Hamilton Baker, MD, FACC, Director, MUSC AI Hub, Medical University of South Carolina.



Engy Fouda

Adjunct Lecturer@Suny New Paltz. Author. Docker,SAS,Kubernetes instructor at ONLC. Harvard University Alumni.Mail list: