Contact us:
040 4016 5703 099 6344 0404
Follow us:

Data Preparation: The Key Ingredient In Efficient AI Algorithms For Virtual Healthcare

There's a saying that a messy kitchen is a happy kitchen. However, that concept doesn't apply to data processing. Artificial intelligence (AI) and machine learning (ML) can't properly execute without good data preparation.

That's an oversimplification, but it helps explain what's going on with the whole ecosystem. Virtual healthcare promised to take healthcare to patients. The fast-growing monitoring and sensing technology in virtual healthcare created explosive growth in available real-world patient data. Data can be tapped to gain valuable information regarding a patient's well-being. However, it's a daunting task to interpret such a large volume of data. AI and ML algorithms are some of the ideal methods to gain insights from such data. Ideally, a patient can live normally and be monitored 24/7 by AI algorithms that send alerts to providers only when an event is detected.

That's an idyllic scenario for virtual healthcare—to use continuous monitoring instead of spot checks to proactively detect changes and expedite providers' ability to provide care. Again, that's a definitive goal, but the reality is that caregivers can be overwhelmed by more information than they can efficiently handle.

To attain the ultimate goal, the provider or researcher must address a major process gap between the raw data and the algorithms. To address the foundation of the issue, start by creating a clear, achievable data preparation strategy.

Look at current processes and the staff required. The time and effort needed to clean data and organize it to be viable for ML applications equate to vast expenditures of manpower—and while it's possible, it can be expensive. InfoWorld references an Anaconda study stating data scientists spend 39% of their time in data prep. That's time that could be spent analyzing data instead of organizing and cleaning it to be evaluated. It can also be a roadblock to progress.

Technology can also organize data so it's ready for AI. This can be a critical component of efficient healthcare, especially virtual healthcare. Providers can make decisions based on algorithm recommendations rather than attempting to wade through loads of data.

Attaining The Optimal System

To return to the analogy, data is food for AI, according to Andrew Ng in a recent Forbes article. However, shopping for ingredients doesn't result in a quality entrée. Those vegetables—like data points—have to be adequately prepared. You can collect the raw information, but it's invaluable until it's prepped and properly combined. Just like a sous-chef is vital to the restaurant kitchen, the right data platform is critical to ensuring data is accurately organized and prepared to make it valuable.

The question becomes whether to build or buy. You must evaluate whether it's more beneficial to focus on existing core competencies or add resources to build data capability as a new competence. Large organizations may be able to afford a new division, but smaller companies usually can't take on developing capability for AI and data analytics—especially since it's not a single project. It's an ongoing process. Aside from adding qualified headcount, there's business and technical risk even for large organizations because the solution must deliver on both capabilities, which requires understanding the IoT and cloud processing technology as well as the sensors. Many companies have opted to take on the risk of building their own enterprise solution—whether it be an email system, database or search engine—and then later abandoned that approach.

On the other hand, choosing an existing platform could require additional capabilities and definitely requires training. When selecting a platform, you'll need to make sure it already has expertise on all of the regulatory requirements (e.g., HIPAA, GDPR, ISO 27001) as well as biometric data processing, IoT and modern cloud data infrastructure, including data warehousing, data lakes, etc.

Attaining The Ideal Mix

A challenge for virtual healthcare is universal data assimilation. A research or provider organization must think beyond attaining information to implementing a means to automatically procure different kinds of sensor data, EHR data and so on instead of doing it manually. Then, that raw information must be processed so it's viable for algorithm ingestion. Doing so ensures that as much information as possible is efficiently garnered from the available data.

To bring our analogy full circle, it's great to grow impressive vegetables. However, if you can't harvest and use them, they don't have value. The devices and methodology to collect high-quality data provide the foundation, but without the ability to act on that information, there's no benefit to the ultimate goal of improving healthcare, facilitating a clinical trial or expediting a drug approval. It's critical for technology to not only ingest high-quality data from different devices but also properly process and organize it to expedite the massive demand for AI development and deployment.

The issue is further complicated by the need to have people who understand human biometric data to process it. It's not as simple as software engineers fitting data into tools or collecting sensor data and building a cloud platform. Both are necessary to establish a robust system. Even if the work is done internally, an IT department must manage all of the externally licensed infrastructures, including navigating regulatory challenges. Success entails understanding the sensors, the wireless communication protocols, the cloud data infrastructure and incoming data processing.

To illustrate, anyone can have a piece of fresh fish, but it takes skill to cut it into sashimi. If you don't properly process the ingredients (the data), you end up with an inedible mess instead of an incredible entree.

You can opt to prepare and cook for yourself, or you can engage an expert chef to assemble the ingredients and prepare the meal. In either case, you must know what you want—whether it's home-cooked (in-house development) or a restaurant (turnkey solution).

No Comments Yet.

Leave a reply