Advances in computer science have paved the way for researchers to accurately diagnose medical conditions, particularly in their early stages. This is a boon for medical professionals, potentially aiding them in planning an effective disease management strategy. In its bid to aid in the quality of these medical treatments, I²R has formed close collaborations with clinician-scientists from the Singapore Eye Research Institute (SERI) and the National University Hospital (NUH) to utilize the potential of computer-aided technologies in the diagnosis of prevalent medical illnesses.

Eye (Ocular) Image-Based Automatic Diagnosis

Aging-associated ailments such as cataracts and glaucoma are the major causes of blindness worldwide. The most common of all ocular diseases is cataracts, with 47.8% of global blindness being attributed to this particular condition. (WHO Report, 2004). Cataracts are mostly related to aging and are common in older people. A community survey was done in Singapore and the results have shown that 35% of Singapore Chinese over the age of 40 years old have cataracts. It has also been uncovered that by the age of 80, more than half of all Americans either have cataracts or have had cataract surgery.

Cataracts are diagnosed as the clouding of the lens in the eye that severely affects vision. There are two ways to diagnose the disease. The conventional method sees ophthalmologists utilizing clinical grading or subjective systems. This is done by comparing the observed lens image to a reference image set in order assign a grade for the stage of the disease. The second method is known as the grader’s grading or objective system. Here, experienced human graders assign a grade that best reflects the severity of cortical opacity based on photographs and digital images so as to classify the lens opacity more objectively.

Although the grading system should have a high degree of objectivity, it has been reported that the measurement of opacity varies between graders and even from the same grader. Furthermore, the measurement of the area of opacity is time-consuming as well. To combat these problems, we have developed an automatic opacity detection approach to grade cortical cataract more objectively. This technology makes use of innovative image processing techniques to achieve and accuracy of 86.3% as referenced to the grading performed by human graders.

Glaucoma is the second leading cause of blindness, accounting for an estimated 12.3% of visually-impaired people globally (WHO Report, 2004). In Singapore, the 2002 Tanjong Pagar community-based study has revealed that glaucoma was found to be prevalent in 3.2% of Singapore Chinese aged 40 years and above.

Glaucoma is physiologically described as the degeneration of optic nerve cells, characterised by changes in the optic nerve head and visual field. Although glaucomatous damage is irreversible, early detection and medical intervention can effectively slow or halt the progression of the disease. Ophthalmologists use the ratio between the optic cup and disc, also known as the CDR, as an indicator of glaucoma.

Currently, CDR evaluation is performed manually. However, this is highly subject to the experience and training of the ophthalmologist. Furthermore, its dependence on manual grading makes conducting large-scale population screening for early glaucoma detection a demanding task. In response to these challenges, I²R developed the ARGALI system. Named after the Central Asian Mountain Sheep that is known for its razor sharp eyesight, ARGALI is a framework which first extracts the optic cup and disc using segmentation techniques, such as level-set algorithms. This is followed by machine learning techniques that will optimally combine these results by automatically calculating the CDR from retinal photographs.

The ARGALI system has been shown to achieve a CDR accuracy of more than 90% within intra-observer limits. It has the potential to provide a fast, objective and consistent measurement that will aid in mass population screenings.

Abdominal Organ Segmentation and Tumour Detection
Liver transplants and cancer detection require good segmentation results for diagnosis. This is especially the case for the treatment of liver cancer and liver failure. For a successful organ transplant, accurate estimation of the organ is especially crucial. However, manual segmentation by doctors is time consuming and expensive, and this has motivated research efforts in the employment of automatic or semiautomatic organ segmentation methods. Ensuring the reliability of liver computed tomography (CT) segmentation and cancer detection is extremely challenging due to the noise present in liver CT images as well as occlusions caused by other nearby organs such as the heart and stomach.

To meet these challenges, I²R is working together with NUH and the National University of Singapore to develop methods based on statistical knowledge extraction from a large liver CT image database. The resulting software will be able to create 2D and 3D mathematical models of the liver and other surrounding organs based on the CT scans, aiding intervention planning and prediction of treatment for liver failure and liver cancer.

Furthermore, these mathematical models can be integrated as "plug in" to currently available clinical medical image systems used in modern hospitals, such as the common PACS (Picture Archival and Communications Systems) platform. The team at I²R is confident that these tools will assist doctors in improving treatment planning for the liver in the future.

Organ transplants and cancer detection require accurate measurements for diagnosis. This is especially the case for the treatment of liver cancer and liver failure. Accurate measurements are crucial in ensuring a successful organ transplant, while inaccurate calculations can lead to liver failure or even death after treatment.

Measurements of the liver are done through the use of blood vessels as their presence and form define the liver segments. However, the CT scans used for this purpose are static and cannot be modified or deformed. This is due to its 2D images, which results in the poor visibility of the structures, making it difficult to fuse them into one 3D model for a full liver and blood vessel view. To combat these limitations, the I²R is currently developing robust software for routine hospital use. This software will be able to create 3D mathematical models of the liver and its vessels by extracting images of the blood vessels and organs from the CT and MRI scans. This would thus result in accurate planning and prediction of treatment for liver failure and liver cancer.

The next step is to programme these software tools to “plug in” to available clinical image viewing technologies so that they may be used in hospital for testing. The team at I²R is confident that these tools will aid doctors in improving the treatment planning for the liver. Meanwhile, the I²R team will continue to develop new health care software for other diseases and organ treatments.

For enquiries or explore collaboration, please contact:
Industry Development Department
Tel: 65 6874 8399
Fax: 65 6775 9923
Email: inddev@i2r.a-star.edu.sg