Screening system for inspecting a patient's retina.
The invention relates to a screening system for inspecting a patient's retina.
Diabetic eye disease (eye disease caused by diabetes) is the most important cause of blindness in the age group 18-74 years. Since blindness can be prevented if treated at an early stage and diabetic eye disease being detectable at an early stage by screening, the World Health Organisation (WHO) and the majority of national health authorities advise yearly ocular screening for almost all patients. Diabetic eye disease affects about 10% of all diabetic patients. In addition, about 8% of all patients above 65 years of age also have diabetic eye disease before developing diabetes .
Due to the early stages of diabetic eye disease be- ing painless and symptomfree, to the large number of patients and/or the high cost of screening by an ophthalmologist or other eye care specialist and to the lack of accessible eye care in large parts of the world, it is estimated that more than 50% of all patients with diabetes are not screened for diabetic eye disease.
The invention aims to provide an alternative to the existing method of manually determining the presence or absence of diabetic eye disease and to obtain advantages that will become apparent from the discussion below. The screening system for inspecting a patients retina according to the invention comprises a retinal imaging device to retain a digital image of the retina, an interpretation system linked to the retinal imaging device for receiving and interpreting the digital image of the retina and an output-device for showing diagnostic information received from the interpretation system.
Surprisingly it has been found that examination of retinal photographs by an ophthalmologist offers higher quality of screening and prevention of diabetic eye disease as compared to manual examination by an ophthalmologist. With the system according to the invention automatic analysis of
retinal photographs is possible offering at least the same quality of detection of diabetic eye disease as compared to manual examination of such photographs by an ophthalmologist.
Advantageously the interpretation system of the screening system according to the invention comprises a detector to detect at least one aspect of the retina that is indicative for diabetic eye disease.
Preferably the detector is suited to detect changes in the retinal image that are indicative for diabetic eye disease.
Suitably the interpretation system comprises a classifying organ connected to the detector to classify signals therefrom.
In a preferred embodiment the classifying organ as- signs a probability value as to the presence of diabetic eye disease in connection with the detected aspects that are indicative for said disease.
The screening system of the invention can operate as a stand-alone system. It is preferred however that the inter- pretation system, the retinal imaging device, and the output device are connected to a data processing network. This provides the advantage that the interpretation system of the screening system according to the invention can be shared by many retinal imaging devices and output devices at different locations.
Advantageously the data processing network is the Internet and the interpretation system is remotely located. Using the Internet will provide a virtually instantaneous and cost effective worldwide access to the interpretation system which may benefit from information gathered throughout the world to improve its analysis capabilities.
The retinal imaging device that is part of the screening system according to the invention is selected from a group comprising a digital camera, a retinal camera linked to a computer, and an analogue camera linked to a scanner.
The invention will now be described by way of a non- limitative example of a preferred embodiment of the screening system according to the invention, utilising the Internet as
the data processing network to which the interpretation system, the retinal imaging device, and the output device of said screening system are connected. It is to be understood however that the screening system according to the invention may also be embodied as a stand-alone system.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT.
Relying solely upon the patient's access to a physi- cian or other health care worker with a retinal photographic camera or other retinal imaging device and a connection to the internet, the present invention provides an accessible, low cost and high quality screening system for diabetic eye disease via the Internet, or an alternative telecommunica- tions network.
All that is required is for the camera operator to transmit the retinal photographs via the internet to a remote site, at which the interpretation system is located. The interpretation system will almost instantaneously provide the camera operator with information on the presence or absence of diabetic eye disease, the probability of such disease, the quality of the photographs, the need for re-screening within a specific period, the need for referral to an ophthalmologist, the presence of other eye diseases and other like in- formation. The software to transmit the photographs and receive the above information is preferably available via the Internet from the remotely located interpretation system.
The apparatus of this preferred embodiment may comprise a plurality of local systems, connected through the internet or other telecommunications network to at least one remote interpretation system for the actual diagnosis to be performed.
The local system offer means for acquiring and transmitting digital retinal photographs to the remote system and for receiving, displaying and interacting with the diagnostic information returned by the remote system.
The remote interpretation system offers means for receiving and storing digital retinal photographs, for the
automatic interpretation of these photographs and for transmitting the diagnostic information to the local system or systems .
The local system is meant to be used as an inter- face from the operator to the remote system, so that photographs acquired at any location (by any means in principle) can be offered to the remote interpretation system through the Internet or other telecommunication network.
It includes a means to acquire digital photographs . This means can be either in the form of a retinal camera, connected to a computer connected to the internet that can convert the photographs to digital form, or in the form of a digital camera that can acquire digital photographs directly, or even in the form of an analogue film type retinal camera and a computer connected to a scanner, so that analogue photographic prints can be scanned in digital form and transmitted to the internet .
In addition, it may offer a computing means so that data including the digital photographs and other patient or photographic information, if any, can be transmitted to the remote system, in compressed form.
In addition, it may offer software performing a user interface for the display of and interaction with the diagnostic information that is obtained from the remote in- terpretation system. This computer will preferably run software automatically or manually downloaded from the remote interpretation system, so that it is always the most up to date version.
In addition, the software may also allow the dis- play of and interaction with diagnostic information received from the remote system, in either textual or graphic form or both. Such information may include, but is not limited to:
1. The presence or absence of diabetic eye disease in the photographs of the patient, photographs which had been transmitted to the remote system beforehand.
2. The probability of the presence of diabetic eye disease in these photographs
3. The severity of such disease, if any.
4. The optimal interval, if any, between this examination and the next.
5. The need for treatment and the type of treatment advised, if any.
6. The quality of the photographs, and advice on how to improve the quality of the photographs.
7. The presence, type and severity of eye diseases other than those related to diabetes, if any. The remote interpretation system (there may be a plurality of such systems) includes means to receive, store and edit digital photographs and other information on the patient or photographs, if any, in either compressed or uncompressed form. This system can be in the form of a computer connected to the internet, with a local or remote storage facility. The storage facility may encompass hard-disks, tape- streamers, compact disks or magnetico-optical disks.
In addition it offers means to transmit the diagnostic information and other interpretation data, if any, in compressed or uncompressed form, over the internet to the local system from which the relevant digital photographs were obtained. In addition it may offer means to translate such diagnostic information into various formats and human languages, such as English, French, Spanish, Dutch and German, and any other language.
The interpretation system is preferably designed to automatically analyze and filter the digital retinal photographs and also determine the presence, and probability thereof, if any, in these photographs of changes that indi- cate the presence of diabetic eye disease.
It includes analysis software to perform the detection of blood vessels, the optic nerve head, the central part of the retina, and the orientation of the retina.
In addition it may include analysis software to de- tect changes in the retinal image known to indicate the presence of diabetic eye disease. Such changes typically consist, but are not limited to, the appearance of white- yellowish spots on the retinal image called exsudates, of
reddish-brown spots on the retinal image called hemorraghes, of sausage-like and spot-like changes in the shape of blood vessels (beading and micro-aneurysms) , and the appearance of new blood vessels called neovaεcularizations . The analysis software performs this classification by image processing computations, such as by an expert system, a neural network or a (plurality of) image filters, or a combination thereof.
In addition the software may make use of clustering algorithms, linear and non-linear mapping algorithms, and pattern recognition algorithms, either individually or as a combination thereof, to prepare the information present in the retinal photographs .
Preferably the system includes diagnostic software that not only classifies the retinal photographs, using the results obtained by the analysis software, in terms of presence or absence of disease, or any particular disease, but also may assign a probability of detection and/or a numerical value indicating the severity of that disease. This provides a tool for monitoring disease progression. The diagnostic software may be in the form of a single-level system or a hierarchical system, both of which may be implemented as one or more expert systems, neural networks or image filters, or a combination thereof .
Advantageously the system includes analysis and di- agnostic software having training module (s), that allow the software to "learn" new or updated diagnoses, and to improve the analytic or diagnostic process. A set of retinal photographs that is stored by the remote interpretation system and suitably selected, edited or modified can be used for the purpose of creating or augmenting a training database. In addition, retinal photographs with empirically or semi- empirically determined or simulated diabetic eye disease changes can be added to this training database. Such a database may also contain only those features, regions or primi- tives of retinal photographs that are related to the detection of diabetic eye disease, instead of containing the whole retinal photographs. With the huge training database thus obtained and continuously updated, leading international ex-
perts on diabetic eye disease can further be employed to improve the accuracy of the entire system.
The construction of a proper training database supports the performance of the interpretation system, including good sensitivity and specificity. As follows from the above, the training database may contain all, or any, of the following types of data: * empirical data, i.e., data obtained from patients with and without diabetic eye disease; * semi-empirical data, i.e., data obtained by modification of the empirical data, as described above, by: •emphasizing or de-emphasizing certain aspects of the image test to bring out the characteristic features of certain diseased states "adding noise or measurement uncertainty of features which may be associated with a real retinal photograph; •any other modification of the retinal photograph, or features, regions or primitives thereof and their associated classification; and, * simulated data, i.e., data that are constructed to simulate the real-world results of a visual field test for both normal and abnormal visual fields.
The interpretation system may have the following modules : 1. a prefiltering module which may filter the retinal photographs over various dimensions such as scale-, color-, clustering- or other spaces;
2. a clustering/data reduction module, which may employ ridge detection, edge detection, shape detection, singular value decomposition, principal component analysis, learning vector quantization, or other clustering or data size reduction methods, to obtain features and other primitives,
3. a normalization module;
4. a classification module, which performs pattern recogni- tion, classification, and quantification of the normalized features and/or other primitives obtained in the clustering/data reduction module through non-linear or linear mapping. This function may be accomplished through the use
of multilayer perceptron neural network or other neural network architectures, or through non-linear, multivari- ate, or linear regression of the data, or by multivariate statistical classifiers or discriminators (such as Baye- sian classifiers) , or by an expert system, or by another such classification system; 5. a training and correction module, allowing the training and improvement of the system through the training database; 6. output module, converting the results from the classification module into an information format suitable for transmission to the local system.
While the particular invention as herein shown and disclosed in detail is fully capable of obtaining the objects and providing the advantages hereinbefore stated, it is to be understood that this disclosure is merely illustrative of the presently preferred embodiment of the invention and that no limitations are intended other than as described in the appended claims .