Biomedical Image Analysis as a Diagnostic Tool in the Clinic

Nabeel U. Ali1,2

1Second-year MD candidate, Albany Medical College, Boston, MA, USA
2Radiology Research Fellow, Harvard Medical School, Boston, MA, USA
Address correspondence to: nabeel.u.ali@gmail.com


Photo Credit: David Blackwell. via Compfight cc.

Photo Credit: David Blackwell. via Compfight cc.

Biomedical Image Analysis is a well-established field of research that is beginning to emerge as a potentially fruitful diagnostic tool in clinical medicine.  Advances in imaging instrumentation have allowed us to visualize the natural world with greater clarity than ever before.  In the laboratory, we utilize imaging to evaluate biological phenomena at the molecular, micro, and nano-scale to derive an understanding of underlying mechanisms.  In the clinic, we utilize imaging to evaluate gross anatomy and diagnose various forms of pathology.  Three imaging modalities in particular, Computed Tomography, Magnetic Resonance, and Ultrasonography, have revolutionized the diagnostic capacity of clinical medicine.  At present, image interpretation by a physician is largely qualitative and occasionally supplemented by manually-acquired quantitative measurements.  Advancement in scientific computation has vastly expanded image-analysis capability, and opened the door to automated diagnostic functions yet to be developed.

Patient images contain a wealth of information in the form of objects and patterns.  The crux of image analysis is the conversion of this image data into quantitative information.  A computer is able to derive meaning from these image features by exploiting morphological, color, and intensity metrics.  Image analysis generally follows a universal sequence: image acquisition, artifact mitigation, feature extraction, measurement / thresholding, and clinical correlation.  Utilizing this process, computational image analysis provides a way to answer meaningful clinical questions.

Computerized image analysis offers several advantages over human vision analysis, mainly in the form of quantitation, automation, and reduced need for human intervention.  Qualitative human evaluation is subjective and not standardized, which results in a relatively low inter-observer reproducibility in reading patient scans.  Utilizing a quantitative method such as computerized image analysis allows for standardized, repeatable analyses able to be performed by multiple clinicians.  This process decreases likelihood of human error-associated inaccuracy.

Another advantage is automation: computerized image analysis is an automated process which exploits the computer’s inherent superiority in rapid calculation and handling large volumes of data.  This is especially important in three-dimensional analysis whereby data is abundant and analysis can be complex.  Image processing time is dramatically reduced which allows greater patient volume handling and an overall improved clinical workflow efficiency.  Further, computerized image analysis automates tasks typically conducted by a human.  Imaging “pre-reads” (any form of superficial or first-pass interpretation of a medical image) may now be conducted by a computer instead of a clinician.

Automated image analysis and computer aided detection as a research domain is in its youth, but promising studies have emerged relatively recently.  Examples from the literature include studies that have proposed methods for automatically detecting micro aneurysms from digital fundus images [1], masses and micro calcification clusters in mammography that are precursors to breast cancer [2], thyroiditis in ultrasound images [3], and cardiac ischemia in myocardial perfusion CT images [4].

Fu et al. describes a computer-aided diagnostic system to classify colorectal polyps by type in colonoscopic imaging [5].  The technology utilizes image-analysis techniques such as component transform, texture analysis and support vector machines to differentiate polyps.  The classification performance of the proposed technology outperformed conventional visual inspection (evaluation by a physician).  This work is a prime example of how image analysis can extend beyond detection functions and provide meaningful clinical insights – potentially better than human analysis.

Segovia et al. proposed a technique for automatic diagnosis of alzheimer’s disease utilizing both PET imaging and neuropsychological test data [6]. Prior work in this research domain has established computer aided diagnostic methods for dementia utilizing image analysis exclusively.  The team showed that in all test cases, the accuracy achieved using both neuropsychological scores and imaging data was substantially higher than the one obtained using only the imaging data.  This study is particularly impactful as it shows the breadth of clinical application for image analysis techniques, especially when used in conjunction with other clinical data.

Looking forward, diagnosis of cardiovascular disease is an effort particularly amenable to developments in image-analysis technology.  There is a large body of research around Coronary Artery Disease (the number one cause of human mortality) that deals with characterization of high-risk plaques as they appear in various imaging studies.  Harnessing this knowledge in order to detect vulnerable plaques prior to a potentially fatal heart attack would prove impactful from a public health perspective.  Application of image analysis techniques to cardiovascular disease is currently an area of intense research and should be supported further.

Today’s critical healthcare problems require a multidisciplinary approach to yield innovative solutions.  Biomedical image analysis is a field of research which applies engineering principles to the biomedical sciences and medicine.  Automated image analysis technology has potential to improve clinical workflow efficiency, tighten diagnostic accuracy, and reduce clinical resource utilization and their associated costs.  In the future, image analysis paves the way for automated diagnosis that may eliminate physician involvement altogether.  Whether or not the algorithm’s output is the definitive diagnosis remains a point of contention, so for now, this type of technology serves to guide physicians.  When image analysis algorithms develop into more clinically robust tools, we must be ready to implement them into our diagnostic protocols and discuss their value for patient care.


References

  1. Kande GB, Savithri TS, Subbaiah PV (2010) Automatic detection of microaneurysms and hemorrhages in digital fundus images. J Digit Imaging 23: 430-7.
  2. Reiser I, Nishikawa RM, Edwards AV, et al. (2008) Automated detection of microcalcification clusters for digital breast tomosynthesis using projection data only: a preliminary study. Med Phys 35: 1486-93.
  3. Koprowski R, Zieleźnik W, Wróbel Z, Małyszek J, Stępień B, et al. (2012) Assessment of significance of features acquired from thyroid ultrasonograms in Hashimoto’s disease. Biomed Eng Online 11: 48.
  4. Liew G, Ali N, Do S, Petranovic M, Cury R, et al. (2013) A novel analysis algorithm for potential quantitative assessment of myocardial computed tomography perfusion. Acad Radiol 20: 1301-5.
  5. Fu JJ, Yu YW, Lin HM, Chai JW, Chen CC (2014) Feature extraction and pattern classification of colorectal polyps in colonoscopic imaging. Comput Med Imaging Graph: doi: 10.1016/j.compmedimag.2013.12.009.
  6. Segovia F, Bastin C, Salmon E, Górriz JM, Ramírez J, et al. (2014) Combining PET Images and Neuropsychological Test Data for Automatic Diagnosis of Alzheimer’s Disease. PLoS ONE 9: e88687.