Case Study: Psychiatric online evaluation system
--------------------------------------------------------------------------------------------------------

Background

Psychiatrists all over the world need to collaborate in order to get the best diagnosis for their patients, because patterns that appear in one set of patients can be matched and applied elsewhere, after a reasonable comparison of symptoms. Today, many hospitals are linked together with an intuitive solution provided by Agni Software – the solution takes care of consolidation of voice and data, collation of patient history and prognosis support from an automated intelligent data analyzer. The solution is built using Borland Delphi, and allows doctors all over the world to interact using a standard PC and Internet connection.

The Challenge

In the field of psychiatry, correct diagnosis depends on correctly assessing a patient, his/her history and current symptoms. There are standard question sets that are designed by organizations such as the UN to allow for an inference-based methodology. Evaluation can also be collated and a better understanding of subjects reached, by automating the process of questioning and rating. But there are hurdles to this approach:

The proposed solution would need to:

Apart from this, the need to train users to use the system was important, especially to the students.

The Solution

Agni Software built a scalable, integrated solution that allows doctors or students to connect to a federalized resource over the Internet or an Intranet. The “rater” conducts an interview, being led through the process by the software, which configures the path of the questions according to the rating given. In addition, the interview is “recorded” in both voice (from a microphone) and ratings, and are transferred to the server in chunks. As the rater proceeds, he gets a current diagnostic tree which displays the potential diagnosis of the current subject, including eliminated parts and inferences.

Architecture

The solution uses Borland’s Datasnap technology (formerly MIDAS), was written in Borland Delphi 5, and was built with close co-ordination with the customer.

Datasnap as a technology was very sound, allowing users to connect to the servers using a number of protocols, including TCP/IP. Agni Software enhanced the Datasnap technology by adding custom built components to allow interaction with multiple databases. The product was built using a spiral model, with base features implemented first and additional modules added on with time. The VCL framework architecture was used extensively throughout the project, and in addition, third party tools such as Infopower and ReportBuilder were used to enhance user productivity.

The architecture was designed to run over the Internet, collating data. The team built custom solutions to compress the data effectively before transmission. Transmission of voice was done in the MP3 format, a revolutionary way to transfer recorded voice. Apart from compression data transfer was multithreaded and done in the background, so that the users would notice no drop in performance.

As the data was very sensitive, many encryption methodologies were used to ensure data safety. Borland Delphi provides direct access to DLLs, which were custom written for the level of encryption that the customer required.

There was the necessity to work “offline” – without any of the data being transferred to another machine. Primarily meant for users who were on the move, or did not have access to an Internet connection, it was also meant as a means for doctors to have private information stores. The Datasnap (MIDAS) technology allowed the use of a briefcase model, where data was stored on the client machine. When the user linked back to a network, the user could synchronize with the network and transfer his stored data if he chose to.

Users in different countries and cultures could “switch” languages without leaving the application – all language specific data is stored at the server and downloaded to the client machine whenever necessary.

Diagnostic support was provided at the server, constantly collating data and generating a diagnostic neural tree based on inferences. Base rules were encoded in a special format allowing changes without affecting the application. Rules were built and modified as subjects were rated and rerated, giving a new interview much more statistical data for inference and diagnosis.

The solution was deployed at the customer’s site and other beta sites, with constant collation of data and fixing any bugs that came up. As more users came online the stability of the solution came to light. Datasnap (MIDAS) allowed us to build stateless servers, so that additional machines could be added to support more users without drop in performance. Additionally, use of fat clients ensured the distribution of processing power between the server and client, thereby ensuring a long lasting, stable, evolving and scalable solution for the customer.

Results

The solution provided is scalable, extensible and secure. The ability to define further rules and build a diagnostic tree is an immense help to doctors to narrow down to the correct diagnosis. The recording of voice and its storage helps doctors rerate a subject later, and the software is capable of playing back an interview exactly as it happened.

Multi-language support is very helpful for those dealing with subjects of different cultures, and the effect was first felt in North America with Spanish and French subjects.

The multi tier architecture and the multi-threaded distribution of processing ensures that the user of the software does not feel a performance lag. And the extensive audit trail support ensures that any legal requirements are complied with. It is no wonder, now, that this system is being introduced in a bigger way in hospitals and mental-health institutes all over the world.