Imaging informatics

From Clinfowiki
Jump to: navigation, search

Imaging informatics is a subspecialty of Biomedical Informatics. The American Medical Informatics Association (AMIA) defines biomedical imaging informatics as a discipline that focuses on improving patient outcomes through the effective use of images and imaging-derived information in research and clinical care.

Its genesis began in radiology and they have become the leaders in Imaging Informatics. Imaging informatics also encompasses many other specialties in which imaging is a key component of the clinical decision making process such as cardiology, pathology, dermatology and ophthalmology.

Two organizational efforts in the 1980’s made it possible now for any radiologist in the developed world to view radiological images from a computer anywhere from most radiological devices. They are DICOM (Digital Imaging and Communications in Medicine) and PACS (picture archiving and communication systems) Picture Archiving and Communication Systems (PACS). Prior to this, it was virtually impossible for anyone to decode images from radiologic imaging devices except for the manufacturers. It was the cooperation of industry and medicine that allowed this successful alliance. In 1983 the DICOM standards committee was formed from these joined forces.

DICOM is a standard for handling, storing, printing and transmitting medical images. It was originally developed by the American College of Radiology (ACR) and the National Electrical Manufacturer’s Association (NEMA). The standards developed were mainly limited to radiology and cardiology, but they are continuously expanding. DICOM allows not only the image but other metadata to be transmitted simultaneously. For example, patient ID and other attributes are also accompanied with the image; information that accompanies the image can never be separated from this image by mistake. There has been near universal level of acceptance amongst medical imaging equipment vendors and healthcare IT organizations of the DICOM standard.

PACS is the interface that allows the DICOM standardized images to be viewed. It is a group of workstations, servers and archives tied to imaging modalities (CT, MRI etc) that acts as a database and processor for these files.

Radiology is an inherently technology driven specialty so it was not unexpected that this specialty quickly saw the need for standardization and coordination. With the explosion of digital images across not only the practice of medicine but in all aspects of society – analog film is becoming almost obsolete. Advances in Imaging Informatics are being tested and applied in other areas of medicine.

Automated Analysis of Medical Images

With the ever increasing volume of digital medical images generated, there is a greater interest in automating the analysis of these images to detect and diagnose diseases. Computer-aided detection (CAD) systems have been developed over the past few decades. The first such CAD system was developed to detect breast lesions and micro-calcifications in mammography. The motivation for such systems was to minimize errors and improve overall accuracy and efficiency of interpreting images.

Application areas for CAD:

1. Clinical decision support systems

2. Imaging based endpoint detection and disease biomarkers in clinical research trials

3. Image analysis in large scale epidemiological studies

4. Screening systems

5. Tele-medicine

Analysis Process and Techniques:

The image processing process can be categorized into four major components: preprocessing of the image, segmentation, region of interest (ROI) analysis, and determination of whether detected structures represent true lesions.

Feature extraction processes are based on

-Known clinical features of the disease/condition: where algorithms are used to identify these features in the image.

-Non-clinical feature based approach: which is data driven, based on image properties such as color, texture and gradient, and not dependent on known clinical feature the disease.

In recent times, CAD systems typically utilize machine learning algorithms for image analysis, where the system is trained using example cases and then develops predictive data-driven decisions about new cases. Another recent advancement is in the use of neural networks to analyze images for disease detection and in certain cases to provide differential diagnosis.


Although deep learning offers much promise with regard to improving the performance of computer aided image analysis, one of the limitations of this approach described by Lee et al. (2017) is the need for huge labelled data sets that are required for training these algorithms. The use of large classified data sets improves the accuracy of these techniques. However, access to medical data, including imaging data, are highly regulated to protect patient privacy. Additional challenges arise from data integration (across data types and systems) and data access issues.

Imaging Informatics in Dermatology

With regards to dermatology, the easy availability of digital photography and better resolution has allowed frequent photography of clinical lesions, dermoscopy, and histopathology. There are currently no standards for cataloguing and storing the files. Often, they are haphazardly stored in various physical locations under various file formats in random organizational fashions. This non-standardization is a concern for privacy and security issues and also for patient care. These images, along with relevant information (metadata) needs to retrieved easily. Relevant information should include, but are not limited to: patient name, date of service, diagnosis and location of lesion. Furthermore, for teaching purposes, this provides for a visual teaching tool/database/atlas. With the widespread implementation of electronic health records (EHRs), the clinical workflow and use of digital images embedded into the medical record is also becoming more widespread. It remains to be seen whether dermatology will unite with the DICOM standards and use the PACS system to access the clinical digital images or pave a different path.

Special considerations for Dermatology include

1) Dermatology is a uniquely visual specialty and diagnosis of a medical condition weighs heavily on what is seen on the patient and what is seen histologically should a biopsy be done. Due to its heavy reliance on the physical exam, photos that are taken can be potentially contain sensitive information and be exploited Detection and management of pornography-seeking in an online clinical dermatology atlas.

2) Dermatologists need to access clinical images for patient care AND also for teaching purposes. Because Dermatology is so visually dependent, the clinical images used for patient care are often used to teach dermatology trainees. These teaching files that were typically kodachromes now exist in the digital realm and the distinction between teaching files and patient records become blurred. Frequently accessed teaching databases include Visual Dx and Derm Atlas There are also commercial ventures that offer software for cataloging images;,

3) An efficient clinical workflow needs to be developed to move the digital images from the camera device into a secured database. With wireless technology and digital cameras embedded in many mobile devices, it seems only logical that digital images should be captured easily at the patient bedside with the appropriate metadata and sent to a secured database without a third party required. Whether this occurs under the DICOM standard or through the PACS system remains to be determined.

Imaging Informatics and CAD in Ophthalmology

Imaging data is most extensively used, researched, and is of much interest in ophthalmology. Imaging is an integral part of ophthalmology practices and provides critical information required for disease diagnosis, to monitor progression and to guide formulation of treatment plans. Similar to other image heavy medical specialties, computer aided image analysis can facilitate the clinical decision making process and provide a way to measure the disease features viewed on the image. In addition, such quantitative assessment and documentation of the disease process would be valuable in determining disease progression and might also influence clinical decisions regarding the best care for the patient.

Significant amount of work has been performed in the area of computer aided image analysis for the detection and quantification of severity in diseases such as diabetic retinopathy, hypertensive retinopathy, retinopathy of prematurity, age-related macular degeneration and glaucoma [5]. Recently, a deep learning algorithm for the automatic classification of macular OCT images, trained on 52690 normal OCT images and 48312 OCT images with diagnosed age-related macular degeneration, was reported to have an accuracy of 87.63% and an area under the curve of 92.78% [6].

Initiatives and Projects





4. J. P. Agrawal, B. J. Erickson, C. E. Kahn, Jr.Imaging Informatics: 25 Years of Progress. IMIA Yearbook of Medical Informatics 2016; Suppl 1:S23-31

5. Zhang Z et al. A survey on computer aided diagnosis for ocular diseases. . BMC Medical Informatics and Decision Making (2014), 14:80

6. Lee et al. Deep Learning is effective for classifying normal versus age-related macular degeneration OCT images. Ophthalmology Retina (2017). July/August, Vol 1(4), 322-327.

Submitted by (Sowjanya Gowrisankaran)