Understanding a Physician’s Handwriting

Lexicus and Visteon Apply the Lessons of AI and Deep Blue

A few weeks ago, chess grandmaster Gary Kasparov defeated IBM’s “Deep Blue” chess computer and program 4-2. To most people, this should scarcely be news since chess is widely viewed as an activity that requires considerable intellect—when played at Kasparov’s level at any rate. However, Deep Blue caught the public’s imagination by not only offering the Azerbaijan-born world champion respectable competition, but also surprising everyone by winning the first match.

The lesson from this is not necessarily that computers are becoming more intelligent in general, but that a proper combination of hardware and software applied to a specific domain can serve that field surprisingly well. Of course, artificial intelligence (AI) researchers learned this rather quickly, shifting a great deal of effort from Alan Turing’s goal of teaching a computer as if it was a child to imbuing it with enough knowledge to allow it to perform acceptably in a specific problem domain. Likewise, Deep Blue may be a grandmaster in chess, but it couldn’t tell you whether to stop or go at a red light.

Pen computing has followed a similar course, and indications are that it will continue to do so. From the dreams of the great horizontal markets came the more practical and realistic vertical applications. Now, software is making the transition as well, beginning to apply specific advantages to more restricted problem domains. Case in point is a new application called BIGVision Prescription Pad, jointly announced by Motorola’s Lexicus Division and Maitland, Florida-based Visteon Corporation. BIGVision Prescription Pad accomplishes what many have felt is impossible—recognizing a physician’s handwriting, often a messy scrawl that can be hard to decipher even by humans.

Using Domain-Specific Information

BIGVision Prescription Pad enables a physician to handwrite a full prescription including the drug name, quantities, dosage, and other instructions. Visteon, a developer of medical information systems, selected Lexicus’ Longhand technology to decipher the physician’s cursive handwriting. The system, which works with either desktop or mobile pen computers, is capable of printing the completed prescription or sending the electronic form directly to the pharmacy. Visteon plans to integrate this with the company’s BIGVision Clinical Manager patient record system.

As you’ll recall, Lexicus Longhand employs neural network technology together with a dictionary based system to understand cursive, printed, or mixed styles of handwriting. Neural technology, once considered an AI technique, but now appearing in a number of sophisticated products, is ideally suited for matching patterns that are mathematically fuzzy, handwriting being an ideal example.

“The technique for understanding particularly messy writing is to restrict the problem domain to the point where it offers both accurate and useful performance.”

The technique for understanding particularly messy writing is to restrict the problem domain to the point where it offers both accurate and useful performance. With Prescription Pad, this involves tuning the dictionary to work well with a physician’s domain-specific information. As David Greenberg, CEO and Chief Technology Officer or Visteon Corporation, put it: “The core technology relies on constraining handwritten entry fields to field-specific dictionaries to maximize recognition accuracy.”

“When a physician writes a drug name, the Longhand engine seeks matches from a list of drugs. When the physician writes a diagnosis, the engine seeks matches from a list of diagnoses. Visteon and Motorola collaborated on applying the Longhand technology to a physician-specific application.”

Domain-specific dictionaries are associated with each field, and the Longhand SDK allows the developer to choose the appropriate dictionary for each field, offering a great deal of flexibility within complex forms. Greenberg added: “A developer can create applications to permit users to add, edit, and delete entries from the dictionaries.”

Processing the Information

Greenberg explained: “BIGVision is an integrated suite of client/server applications that addresses clinical, operational, and financial processes for large physician group practices.” Built entirely on Microsoft’s strategic enterprise platforms, including NT Server, NT SQL Server, and Visual Basic, the system supports a variety of hardware platforms and communication protocols, as well as healthcare industry standards such as HL7.

Greenberg noted: “As a client/server application, the BIGVision Prescription Pad has access to all system services provided by Microsoft BackOffice, including database transactions, e-mail, direct faxing, printing, and system management.”

In fact, Visteon is betting that it will be able to leverage a number of Microsoft’s leading architectures, including OLE, MAPI, and NT, to bring robust solutions to market faster. Greenberg noted that this will: “enable us to rapidly incorporate technological advances into our core products, without the common constraints that limit the flexibility of many legacy applications.”

BIGVision Prescription Pad is designed to run under Microsoft Windows 3.1 and Windows 95 with the appropriate Pen Computing extensions. Hardware requirements are in line with current Windows applications: a 486 or better processor equipped with 16 MB of memory. According to Greenberg: “Compatible portable devices include mobile computers from Fujitsu, Epson, and others, as well as digitizer/LCD peripherals for desktop computers such as the Wacom PL-100V.”

In terms of developing custom forms, Prescription Pad does not currently include a set of tools allowing a clinic to create or modify forms, however, Visteon noted that BIGVision Clinical Manager release 2.0 will include a “template construction kit”.

“What BIGVision Prescription Pad and similar products illustrate is that workable solutions are achievable once the domain specific nature of the problem is separated from the overall context.”

Breaking Medical Barriers

What BIGVision Prescription Pad and similar products illustrate is that, not unlike grandmaster-level chess and other seemingly intractable situations, workable solutions are achievable once the domain specific nature of the problem is separated from the overall context. Prescription Pad, with its use of specific dictionaries, is a relatively simple example, but one that nonetheless solves the problem at hand. Solutions to other types of problems may also benefit from this general approach.

Looked at another way, the real task may not necessarily be to solve problems as they stand, but to take best advantage of the existing technology and work the problem around the strongest features of today’s systems. In identifying the barriers of pen computing in medicine, Greenberg concurred: “A major barrier is the current high cost of robust pen-computing platforms. In addition, current implementations of pen-based applications often reach out too far in front of the technology, creating a gap between expectations and results.”

Contact:

Visteon Corporation
2301 Lucien Way
Suite 200
Maitland, FL 32751
David Greenberg
(407) 643-7669
[email protected]

Contact:

Motorola, Lexicus Division
490 California Ave., Suite 300
Palo Alto, CA 94306
Brenda Scariot
(415) 833-8062
(415) 323-0482
[email protected]