Real cost savings for telehealth may be potential as published telehealth cost analyses include costs associated with development of the network infrastructure. Our approach has utilized commercial Internet connectivity to assess and validate the application of content-based image retrieval (CBIR) technology and its integration without incurring significant infrastructure costs.

Technological advancements have made remote delivery of consultative health care feasible and widespread in recent years. With the ubiquitous connectivity of Internet, and the increasing availability of network bandwidth, exchanging medical information and delivering services over a large geographic region is possible, while providing high throughput for real-time data analysis. To achieve the goal of automating the detection and diagnosis of diabetic eye disease in real time from digital images taken in a primary care setting, the underlying network infrastructure needs to be established. In ocular telehealth networks, the participating primary care clinics are connected to a diagnostic computer server using standard Ethernet connections. The web-based telemedical network adds portability and significant flexibility in healthcare delivery, providing access to expert diagnosis and high-throughput potential to meet the growing need for disease assessment and management in rapidly expanding at-risk and underserved patient populations. In this chapter, we present the results of our work developing and implementing image analysis and management algorithms that can leverage just such a clinical and retinal image database to improve the sensitivity and specificity of retinal diagnosis in a robust, objective, and deterministic manner using content-based image retrieval (CBIR). In the following section, the design and implementation of a telemedical network are described.

5.2Design Requirements

To provide a network infrastructure for automated diagnosis and analysis in a primary care environment, the security, reliability, and integrity of the data exchange is crucial in the design requirements. To ensure security, the data transmission

protocols must be cryptographic and compliant with FIPS (Federal Information Processing Standard) 140-2, and the patient-sensitive information must be encrypted to meet the HIPAA compliance requirements. To guarantee data reliability, the data flow in the network must be robust under any network connectivity conditions, and the data transmission has to be fast to provide high-throughput image analysis and data processing. To assure data integrity, validation for data entry needs to be enforced, and the data storage in the system needs to be continuously monitored.

5.2.1Telehealth Network Architecture

In the network design for ocular telehealth, a client–server model approach is implemented for the data transmission between the client and the diagnostic server. Undilated fundus images of both eyes are captured by a nonmydriatic commercial retina camera and are exported and submitted to the server for an automated quality assessment (QA) metric. The QA metric assures adequacy of the images for the purpose of determining a diagnosis. The QA results are communicated to the end user in real time. Inside the server, the automatic diagnostic service retrieves the images and defined patient metadata and performs image processing tasks to identify anatomical structure location, extract features, and detect lesions. Following the analysis, the algorithm assigns a diagnosis with an appropriate management plan and generates an encrypted report accessible by the referring physician. An ophthalmologist then reviews the images and the diagnostic report and provides validation before the confirmed report is returned to the end user. An ophthalmologist can also override the computer generated diagnosis by manually assigning a different diagnosis through the web interface to generate an alternative or modified report and management plan.

To provide secure data communications between the clinics and the diagnostic server over the Internet, all data transmission protocols in the network are cryptographically designed to meet the standards published in the Federal Information

Processing Standards (FIPS140-2). To transfer image files to the server and communicate the quality assessment result (pass or fail) to the camera, a Secure File Transfer Protocol (SFTP) is used. To access the report generated on the server, the referring physician can either login on a secure website using HTTPS (HTTP over Secure Socket Layer) or download an encrypted PDF report sent as an e-mail attachment.

To ensure security of patient-sensitive information and to meet HIPAA compliance requirements, data encryption is performed in the reviewing and confirmation process. When the review is complete, the ophthalmologist digitally signs and encrypts the report using an X.509 certificate and then sends the confirmed report with digital signature and encryption to the referring physician in the end user clinic. PDF encryption with an issued certificate guarantees data integrity and security while providing a passwordprotected access of the report for referring physicians via an e-mail attachment.

To fulfill the requirement of robust data transmission under adverse network conditions or unstable network connectivity, the client program on the camera closely monitors the network status and reacts to the connectivity instability in a real-time manner. In the multithread client monitoring program, one of the threads is devoted to scanning the network status constantly and reporting the network status to the user every second. If connectivity issues occur, the client program reacts promptly to attempt self-recovery of network connections, if possible, and prompts interactive information to the user. If self-recovery cannot be achieved within a preset time limit and network problem persists, the client program prompts the user to address the problem. If there is data transmission when network problems occur, the client program retransmits the exported images after the network is reinstated. To guarantee that there is no loss of data in the client–server communication, a persistent acknowledgement is required for committed storage in the bidirectional data transmission.

Reliability and data integrity is critical in designing the ocular telehealth system. Reliability requires the completeness, timeliness, and

accuracy of the data; integrity refers to the validity and consistency of the data. Reliability and data integrity can be compromised in a number of ways, for example, human errors when data is entered, errors that occur during data transmission, viruses, hardware malfunctions, and even natural disasters. To minimize these threats, the following methods are incorporated in the system design: regular backup to another independent physical media in the university server farm and elsewhere; controlling data access to via security mechanisms; designing user interfaces that prevent the input of invalid data; and bookkeeping the critical status of every procedure involving data manipulation and analysis to provide an audit trail of data flow in the system.

5.2.2Work Flow

The work flow of the ocular telehealth network is illustrated in Fig. 5.1. Data storage can be either in file format inside directories or in a database format. The procedural modules include quality check and thresholding, a diagnostic engine running to perform disease stratification for incoming data and to build a CBIR library based on data archival, report generation, and bookkeeping functions such as log and process records.

Three mechanisms for data communication are involved in the network flow. The communication between the remote clinics and the server is through SFTP connections, the report generated is communicated back to the referring physicians in an encrypted e-mail attachment, and the communications between the data storage modules and the procedural modules are through an internal university network. The interface design in the network is composed of three parts: (1) the client application GUI for the remote clinics, (2) the web GUI for the authorized physicians, and (3) the service GUI for monitoring and bookkeeping, which is transparent to the end user. In our web-based network, retinal images from diabetic patients are encrypted and transmitted from the fundus cameras at the primary care clinics to the diagnostic server using DICOM protocols.

Fig. 5.1 HIPAA-compliant telehealth infrastructure and image analysis process. The client application encrypts and exports patient images and metadata to a dedicated diagnostic server. Image processing and analysis procedures are performed in an automated fashion including a quality assessment (QA) to ensure adequacy for anatomical structure analysis, lesion detection, and assignment of an

automated diagnosis according to the posterior probability of each defined disease state using our content-based methods. The server provides an interface for the consulting ophthalmologist to generate a signed diagnostic report which is encrypted and returned to the end-user physician.

The client application GUI provides an interface for the regional primary care clinics for monitoring the data transmission and network status. The clinic acquires fundus images of the retina using VisuCamPro NM (nonmydriatic) camera (Carl Zeiss Meditec). The client application GUI closely monitors the image exportation on the camera. New images are immediately encrypted, exported, and transmitted to the dedicated diagnostic server, along with patient metadata entries from the clinics. At the same time, the client application GUI constantly scans the network status and reacts in a real-time manner to prompt interactive information when connectivity issues do occur.

After the images are submitted to the server, image processing and recognition procedures are triggered on the server to perform automated diagnosis. However, those procedures are transparent to the users. The procedures involved can be briefly described as the following: the submitted images are first subject to a quality assessment (QA) to ensure adequacy for further analysis. Images that pass QA undergo automated processing, including anatomical structure analysis, lesion detection, and automated diagnosis according to the posterior probability of each defined disease state using CBIR methods that evaluate the retrieval response and stratify the disease state, as described below.