Multislice CT and CAD bring new efficiencies to diagnosis

November 1, 2008

Computer-aided detection is gradually gaining acceptance in radiology and has become a major research focus in the past few years. The development of CAD with multislice CT has reached the point where, together, they have the potential to offer new capabilities in the interpretation of emergency room scans.

Computer-aided detection is gradually gaining acceptance in radiology and has become a major research focus in the past few years. The development of CAD with multislice CT has reached the point where, together, they have the potential to offer new capabilities in the interpretation of emergency room scans.

Various versions of the CAD acronym have been suggested, including computer-aided detection, computer-aided diagnosis, and computer-assisted detection. Today, it is universally accepted that any CAD applications will assist a radiologist rather than serve as independent diagnostic tools, and therefore the term "detection" is more appropriate.1 The recognition of virtual colonoscopy for the screening of precancerous colonic polyps and of CAD for generating automated segmentations of liver and kidney has encouraged further work to auto-mate advanced postprocessing applications.2-4 Although the potential of computer-aided techniques for use with MRI data sets has been explored for a few clinical needs, there is more acceptance of CAD algorithms with MSCT.5,6

Novel methodologies and techniques, like CAD, should always be directed to look for solutions to common and routine problems. Emergency radiology accounts for 30% to 40% of all CT scans and is important to patient care because of the frequency of life-threatening presentations in the emergency room. Given the importance of emergency radiology and its potentially crucial impact on standards of patient care, it would be wise to encourage the development of MSCT-CAD applications geared to solving the common problems in diagnosis of acute, life-threatening clinical conditions.

Certain advantages of MSCT are unique to an emergency radiology setting. Generally, a seriously ill or injured patient is likely to be imaged on a modality that ensures good accuracy and reinforces diagnostic confidence of radiologists and clinicians. MSCT has performed the best so far in maintaining the delicate balance between accuracy, readers' confidence, and time savings in the emergency room. MSCT also offers the best diagnostic accuracy for most clinical conditions that present in emergency situations and is preferred over MRI because of its faster scan speed and better image resolution. These qualities also lower the probability of encountering motion artifacts in the images, which is a common problem in an emergency radiology setting.

Research and development that explore the utility of computer-aided techniques has been uneven across the various dimensions of clinical practice. Practitioners in a clinical setting put up with a high rate of false-positives because of the better accuracy and specificity of CAD results. But concerns over false-positive results have delayed exploration of CAD's potential in an emergency radiology setting, where the instant and accurate detection of life-threatening conditions is of paramount importance.

The need for scan data sets conducive to postprocessing and the difficulty of integrating CAD results from a 3D workstation into PACS form the basis for most apprehension. However, the advent of MSCT and the ever-improving resolution of CT data sets have opened the door to development of pulmonary embolus detectors and ways to locate intracranial bleeding sites.

The ability to maximize a lesion's conspicuity and achieve a clear definition of its borders is essential to ensure acceptable accuracy in CAD results. A computer-aided lesion detector can produce optimal results only if the lesion borders are well defined. Fuzziness around the lesion boundaries of lung nodules was routinely encountered with earlier generation CT scanners when sections were obtained at thicknesses of more than 5 mm. This increased the number of false positives not only through erroneous mapping of adjacent vasculature but also by appearing to alter the actual size of lesions. The availability of thin sections with 16-slice and higher scanners has addressed this issue and led to potential solutions.7

Image resolution further influences the clarity of lesion borders. The faster scan speeds available on newer generation CT scanners have dramatically decreased the probability of mo-tion artifacts, thus providing a solu-tion to another problem of the ER setting.

The CAD highlights for emergency radiology that have been displayed in various technical journals and conference platforms have used sophisticated technical terms like level-set and thresholding to describe CAD techniques. The crucial initial step in development of any computer-aided CT lesion detector in most situations is defining a preset range of Hounsfield units such that CAD can display the voxels that fall in the assigned range; e.g., −900 to −930 HU for detection of free air in cases of pneumothorax.

Voxels, which are mapped in color by CAD, are then used to further facilitate depiction of lesion volume and total lesion burden. The same considerations are also the basis for extending CAD concepts beyond lesion detection to facilitate automation in postprocessing.

Another basic principle that is being used to further drive development of CAD applications is the detection of pathologies based on variations in lumen diameter. CT colonoscopy has provided an ideal platform for virtual-navigation CAD, and the same concepts are being applied for detection of transition points of small bowel obstructions.


In terms of research, the exploration of CAD applications in ER is relatively new. A realization of the potential importance of computer-aided principles in emergency radiology is too recent for much data to have been disseminated broadly. Its popularity has been restricted to various conference platforms so far, and data may be waiting in the pipeline for publication.

CAD's potential for detecting intracranial bleeding sites and pulmonary embolism has been established by studies recently published in various journals. However, an interesting array of emergencies in which the new role of CAD is gaining momentum deserves special mention.

• Pneumothorax. The importance of automated detection and quantification of pneu-mothorax received prominent recognition at the 2007 RSNA meeting.8 Assigning CAD to use a range of Hounsfield units on MSCT data sets for mapping free air in the pleura of trauma patients revealed promising accuracy in quantification of pneumothorax and also highlighted the time savings. With computer-aided automated volume-try finding a strong correlation coefficient-nearly 0.9-between the clinical score and manually measured free air in pleural space, promising results are eagerly anticipated.

Results already achieved have encouraged researchers to go a step further and test CAD performance for quantifying pneumothorax in emphysematous patients and work on erroneous mapping of emphysematous bullae present in the lung. Clinicians wishing to ensure clinical stability may continue to place a chest tube.

This may sound discouraging for those attempting to develop CAD's ability to quantify free air in the pleura, but CAD can still play a role in the emergency room for monitoring of a relatively stable patient with a small pneumothorax. Other potential roles include situations just beginning to be explored, such as post-lung biopsy. Here, the idea is to avoid an invasive procedure in an otherwise stable patient.

• Pulmonary embolism. PE is a life-threatening emergency requiring prompt diagnosis to ensure a good clinical outcome. It is the only condition routinely encountered in the emergency setting for which the use of CAD has been extensively investigated.9,10 Newer computer-aided detectors have the ability to automatically map tiny clots, even in the sixth-order branches of pulmonary vasculature. With the recent advances in scanners and scanning techniques, sensitivity has been as high as 85%. Specificity has ranged between 90% and 95%.

The technical learning curve with the improvement of scan resolution and speed has brought down the false positive per patient rate from 4.5 to two. More recent studies have high-lighted the benefits of adding CT venography to regular PE protocols on MSCT, which raises sensitivity to 90%. Some clinicians may still prefer going through the scan data slice by slice, but it seems possible that CAD could be used as a second opinion for conditions such as PE in which care and vigilance in reporting blood clots is important.

• Small bowel obstruction. Use of CAD for locating transition points of small bowel obstructions, such as bowel strictures, was investigated in a 15-patient study by Sainani et al11 that found that CAD identified at least 60% of surgically confirmed strictures with a false positive rate of three per patient.11 Of these, 80% were easily dismissed as erroneous CAD points in vessels, abdominal wall, or collapsed large bowel segments. Considering that redundancy and variability in small bowel position are likely to continue posing challenges, even small cohort pilot studies such as this may point to a solution.


Exploration of CAD's potential in the emergency radiology setting is just beginning. At this vital point of technical advancement, it may be time to focus on CAD research. With the use of terms such as "detection" instead of "diagnosis," a broader meaning that includes automated postprocessing techniques can be included under the definition of CAD.

Patient care in the emergency setting continues to be a challenge. Patients are frequently scanned with CT, and directing CAD research to find ways to quickly and accurately diagnose life-threatening emergencies could have immediate and far-reaching benefits. Improving CAD of thoracic and abdominal problems, which are responsible for a considerable number of emergency presentations, may help reduce mortality and morbidity in the emergency setting.

One can imagine that the CAD concept of shape-based detection, which is used for colonic polyps, may also work for detection of tiny causes of abdominal pain such as a renal stone that may be overlooked by the naked eye. However, false-positive detections due to renal artery calcification and inclusion of aortic calcification foci within the region of interest may pose challenges for improving software accuracy.

CAD colonoscopy that can detect polyps as small as 5 mm in diameter has already paved a path for further application of such concepts. This may encourage attempts to use CAD to locate occult bleeding sites in the small bowel lumen and, if that's successful, may also be used in conjunction with digital subtraction images toimprove accuracy in mapping the bleeding sites.

At this point, CAD's success in locating such occult bleeds may be variable, considering the sites' tiny size, possibly even smaller than 5 mm. But, thanks to such advances in scanners as high-resolution CT, more refined MSCT data may soon be available, which may make generation of such applications possible.

• Musculoskeletal emergencies. It's also possible to imagine developing CAD fracture locators in patients who present with multitrauma. Such an application might have a role in detecting hairline fractures not visible to the naked eye on coronal and sagittal reformats.

A possible danger in this approach is that such CAD algorithms may be programmed to detect subtle discontinuities along the bone's cortical surface that may be confused with artifacts, thus giving rise to false-positive detections. Ensuring exclusion of joints in mapping CAD results may decrease the number of false positives to some extent, but the condition of patients in the emergency setting and the possibility of motion artifacts may pose hurdles and slow research.

• A CAD-conducive research environment. It is important to ensure a good initial groundwork by considering the viewpoints of various schools of thought before formulating a future CAD project. Investigating CAD algorithms on a prospective basis in an emergency setting may provide more reliable answers to queries and misconceptions about CAD and thus help in tapping its future potential.

Setting up a close interaction between an institution's 3D laboratory and the emergency room may provide vital information on some of the common clinical emergencies while we await technical advances in scanner consoles and the integration of CAD and PACS.

The emergency room is a setting that demands vigilance to ensure good patient care and contributes a considerable volume of CT to the radiology workflow. Understanding its importance, and considering the advances that newer MSCT scanners can offer, it would be prudent to add more weight to CAD research in emergency radiology.

With emerging answers for spotting intracranial bleeding locations, tiny pulmonary emboli, pneumothorax, and transition sites of small bowel obstructions, we can hope for a more aggressive approach by CAD researchers to providing diagnostic solutions to emergencies while simultaneously refining CAD's capability to act as a "second reader." Research and development of CAD concepts in busy emergency scanner console and reporting rooms should be promoted to explore the parameters and dimensions of CAD in patient care.

Dr. Singh is a radiology research fellow, and Dr. Yoshida is head of the research group, both in the 3D imaging department at Massachusetts General Hospital in Boston. Dr. Sahani is director of CT at MGH and an associate professor of radiology at Harvard University.