Digital diagnoses—the shape of things to come

March 2007
Feature Story

Karen Lusky

Digital pathology is playing a growing role in helping pathologists expedite diagnoses and, in some cases, make more accurate interpretations. And that may be only the start of digital things to come.

Today, digital pathology and computerized algorithms aid in diagnosis or provide clinicians with information related to treatment. But they may in the future lead to more “directed therapies…where you treat a specific patient’s tumor versus a tumor type,” predicts Keith J. Kaplan, MD, attending pathologist at Evanston (Ill.) Northwestern Healthcare and assistant professor of pathology at Northwestern University Feinberg School of Medicine, Chicago.

As for image-based diagnostic solutions that are available now, Dirk G. Soenksen, CEO of Aperio Technologies in Vista, Calif., a provider of digital pathology systems, is seeing more software using morphometric image analysis. Examples are software for counting positive nuclei for markers, such as estrogen or progesterone receptors, and software that measures membrane completeness for markers such as HER2.

“Morphometric algorithms automate tasks that a pathologist could do manually,” Soenksen says. But they “do so more objectively because a computer can count the cells in a large region more accurately than a human.”

Put another way, “Computers are dumb but accurate,” says Ossama Tawfik, MD, PhD, professor and director of anatomic and surgical pathology and vice chairman of the education and outreach department at Kansas University Medical Center, Kansas City. “If you have brown dots on the screen and ask the computer to count those, it will count them the same way every single time.”

Kenneth J. Bloom, MD, chief medical officer and medical director of Clarient Inc., Aliso Viejo, Calif., says the new CAP/ ASCO HER2 guidelines for optimal HER2 performance (Arch Pathol Lab Med. 2007; 131: 18–43) encourage image analysis in equivocal cases.

“Certainly,” Dr. Bloom says, “when the pathologist is at a decision point, image analysis can provide a more reproducible assessment.” He adds, “The key word is reproducible,” not necessarily more accurate. The pathologist has to determine accuracy as “part of the assay validation procedure as well as confirm the improved reproducibility.”

Digital pathology provides pathologists in any location virtual access to a state-of-the-art immunohistochemistry lab with a full menu of antibodies and image analysis. For example, pathologist clients of US Labs, an AP reference lab in Irvine, Calif., do virtual image analysis. “We perform the stains here and scan the slides using our digital scanning system” and then post the images on the Web for clients to access, says Dennis P. O’Malley, MD, US Labs’ medical director.

Clients select the field or fields they want analyzed and push a button to apply the imaging analysis algorithm. “Each field selected is analyzed and provides a final score,” he says.

The imaging analysis system overall is used most commonly in breast cancer for ER, PR, HER2, and sometimes Ki-67, a proliferation marker. It can also be applied to EGFR for colon cancer, he says. US Labs is now evaluating image analysis for new prognostic markers for breast cancer, prostate cancer, and lymphomas.

Pathologists at John C. Lincoln Hospital, Phoenix, use US Labs to perform IHC panels, including image analysis, for breast cancer, which includes ER, PR, HER2, and Ki-67, says Blair I. Reuben, MD, the hospital’s co-director of pathology.

For all other immunostains, “US Labs provides a digital image to us to interpret as positive or negative,” Dr. Reuben says. An example would be BCL1 for mantle cell lymphoma. “We look at the image the minute they post it on their site and can give the oncologist a positive or negative result,” Dr. Reuben says. A positive BCL1, which indicates a mantle cell lymphoma, can change the patient’s treatment “drastically,” he notes. “A mantle cell lymphoma can masquerade histologically as a low-grade B-cell lymphoma, but it is more aggressive clinically and can be lethal if not properly treated.”

Using virtual IHC, Dr. Reuben’s lab can give clinicians results in 24 hours compared with the two to three days it would take if they waited for the back-up slides from US Labs. Of course, a 24- to 48-hour difference in turnaround time is not going to change patient outcome, Dr. Reuben admits, yet “as a pathologist, you want to have the shortest turnaround time possible in the patient’s best interest.”

Digital capability can pave the way for same-day turnaround of outpatient breast biopsy results.

The University of Arizona’s University Medical Center in Tucson is using the DMetrix Virtual Slide Imaging System to help provide pathology reports by 3:30 to 4 PM for patients who have breast biopsies at a breast center in the morning, says Ronald S. Weinstein, MD. He is professor and head of the Department of Pathology at the University of Arizona College of Medicine and a member of a private pathology practice, University Physician Health Care, that provides AP services to the medical center.

The lab uses a small rapid tissue processor, the Milestone (Shelton, Conn.), for the biopsies. The biopsies are processed in a nearby smaller university hospital where Dr. Weinstein’s pathology group generally has only one pathologist on site. So when a diagnosis is positive for cancer, the lab digitizes the slide using the DMetrix system and transmits the image to a second pathologist to review at the University of Arizona’s University Medical Center.

The DMetrix, which Dr. Weinstein designed along with engineers from the University of Arizona College of Optical Sciences and a spin-off company, DMetrix Inc. of Tucson, generally takes a minute to scan a glass slide with a 1.5 ¥ 1.5 cm tissue section, he says. Once the second pathologist reviews the digitized slide, the report is generated immediately.

Then the oncologist enters the picture—literally.

“When a patient gets a diagnosis of breast cancer at 4 PM, for example, we can link the patient immediately to an off-site oncologist by bidirectional video conferencing…,” Dr. Weinstein says. That way, the patient can have her first clinical interactions with an oncologist to discuss a treatment plan.

“Surgeons say the patients have started their journey with cancer by making the contact with an oncologist and are better prepared,” he says. (A recent publication describes the entire process: Weinstein, et al. IBM Systems Journal. 2007;46(1):69–84; www.research.ibm.com/journal/sj/461/weinstein.pdf)

Dr. Weinstein’s pathology group is planning to commercialize the ultra-rapid breast care service. “In the future, we will be able to do HER2, PR, and ER analysis as part of the quick turnaround laboratory service,” Dr. Weinstein says. But they first wanted to study the impact of this kind of rapid service on their pathology practice, which they have now done. “Our pathologists are very positive about the approach…” So “we are now looking at the incremental advantage of adding special stains the same day.”

Digital pathology can also help standardize diagnoses and prognoses for prostate cancer. “You could digitize the standards for Gleason grading…and make them available on an Internet Web page to facilitate standardized application of grading criteria,” says Jeffrey L. Myers, MD, A. James French professor of diagnostic pathology and director of the Division of Anatomic Pathology at the University of Michigan, Ann Arbor, which is weighing various digital pathology applications.

To help prostate cancer patients navigate treatment options postoperatively, Edouard J. Trabulsi, MD, director of minimally invasive urologic oncology at Kimmel Cancer Center, Thomas Jefferson University, Philadelphia, is using Prostate Px, an integrated molecular test from Aureon Laboratories, Yonkers, NY, that can predict prostate cancer or PSA recurrence.

Prostate Px provides a numeric score and percentage prediction that the patient will experience PSA recurrence or clinical failure after surgery within five years, Dr. Trabulsi says.

How does Prostate Px work? Pathologist Michael J. Donovan, MD, PhD, Aureon’s senior VP for research collaborations, says it incorporates an image analysis system integrated with a patient’s clinical data and biomarker attributes to generate predictive outcome models.

Aureon has “developed a mathematical algorithm that allows us to look at the clinical and morphometric data and immunofluorescent images and build a predictive model that looks at all three domains, identifying a template of features associated with outcome for individual patients,” Dr. Donovan says. (For an in-depth explanation of Prostate Px, see “Will Systems Pathology Replace the Pathologist?” in Nature Clinical Practice Urology. 2007; 4: 39–45 at www.nature.com/ncpuro/journal/v4/n1/full/ncpuro0669.html. The answer to the question in the article’s title is a resounding no, say Aureon pathologists.)

Dr. Trabulsi says he uses Prostate Px for “patients who have questionable pathology after surgery when we are trying to decide whether the person needs further treatment or just observation.” The test can be helpful in deciding whether to consider adjuvant radiotherapy.

Ricardo Mesa-Tejada, MD, Aureon’s VP of pathology and medical director, says that in Prostate Px, nonclinical features associated with clinical failure include, “among others, androgen receptor intensity in various compartments and texture variation in cellular elements.”

Dr. Trabulsi has been using Prostate Px for about eight months and is awaiting Aureon’s Prostate Px Plus, which is expected out soon and which clinicians can order for prostate biopsies before men have been treated.

He believes it will be able to help determine which patients may be suited to radioactive seed implants rather than surgery. “There’s some concern that the treatment [implants] isn’t as effective in intermediate- or high-risk patients.”

Some hospitals are moving toward use of digital pathology for frozen sections. Aperio’s Soenksen says pathologists are becoming increasingly aware that digitizing an entire frozen section specimen for remote access can be beneficial. It offers “performance advantages,” he says, compared with robotic microscope approaches using a video camera to provide a live image of a specimen.

“The major benefit of whole-slide imaging is that any area of the entire digital slide can be accessed at any magnification instantly,” Soenksen says. “Robotic microscopy only provides access to a tiny region of a glass microscope slide, limiting the utility of live telepathology to small specimens. In addition, robotic microscopy is limited to the glass slide placed on the stage of the instrument,” whereas “any number of whole-slide images can be stored and accessed on a server.”

Kansas University Medical Center began a few months ago to use digital for frozen sections in selected cases. The operating room staff sends the abnormal tissue to Dr. Tawfik, who makes a slide, digitizes it, and sends it to the surgeon in the OR via the Internet. Then the dialogue begins. “The surgeon may ask how I’m coming up with a diagnosis and where I got the tissue from. I may tell the surgeon that a specimen showing kidney cancer, for example, is too close to the resection margin and he or she needs to take more normal tissue.”

Looking ahead, Clarient’s Dr. Bloom envisions software that can help the pathologist with pattern recognition. “When confronted with an unusual tumor, the pathologist would be able to push a button like a ‘Google search’ to query all images that the system has ever encountered that had a similar pattern,” he says. The program would then present the pathologist with the images that had a similar appearance and provide him or her with analysis tools—for example, suggesting stains or molecular tests to help differentiate the possibilities.

The software might even suggest that before diagnosing a lesion as neoplastic, the pathologist might consider “other benign entities” that look just like the lesion he or she is examining, Dr. Bloom says. “The pathologist could then say, ‘Wait a minute, I am unfamiliar with entity A or entity B’ and review those entities online directly from the archive.”

Yale pathologist John H. Sinard, MD, PhD, says what may change the utility of digital in years ahead is the ability to “take a digital image and feed it through computer algorithms that will assist” the pathologist in making a diagnosis.

A pathologist “could scan in 10 slides from 10 prostate biopsies…and have the computer identify the five most suspicious glands.” Then he or she “would just have to look at those to say if it’s cancer or not,” says Dr. Sinard, professor of pathology and director of pathology informatics, Yale University School of Medicine and Yale-New Haven Hospital.

Ulysses Balis, MD, director of clinical informatics at the University of Michigan, says he and his colleagues have compelling pilot data suggesting they “can develop an algorithm that is exquisitely sensitive to slight change in morphology that will scan whole-slide imagery for malignancy or premalignancy. It can also scan “for inflammatory-type lesions…so it has uses beyond cancer.”

KUMC’s Dr. Tawfik and his research colleagues have set their sights on using an automated digital pathology solution to help improve grading of breast cancer, which he says hasn’t changed much since the 1950s and about which there are accuracy concerns.

The results of his research have shown that combining an automated count of the proliferation marker, Ki-67, with the tumor’s nuclear grade into a new grading system provides a better correlation with overall survival of breast cancer patients than does the old system.

“By incorporating automation in grading tumors,” he adds, “we could potentially be able to provide more accurate tumor biology and prognostic information, allowing clinicians to better customize treatment for their patients.”

Some say genetics is the answer to determining prognosis, he notes. “But morphologists like me say it’s not that grading is a bad technique, it’s that the method we use to grade cancer isn’t accurate. Let’s improve the method and then talk.”

Also in the digital pipeline: Multispectral imaging, which is “trembling on the verge of being highly clinically relevant,” says Richard Levenson, MD, director of research, biomedical systems, CRI Inc., Woburn, Mass. CRI has a multispectral imaging product, Nuance, that is used now in pathology and drug research applications.

Nuance can be added as an accessory to any standard microscope, he says. The system has “a digital CCD camera equipped with a no-moving-parts electronic filter that can rapidly take images at multiple wavelengths, along with...software with application-specific modules.”

Multispectral imaging—which “is to color imaging as color imaging is to black and white”—allows you to capture and measure cellular markers all at the same time and on the same tissue section, Dr. Levenson explains.

Using Nuance, the pathologist can look at one slide and identify several different markers simultaneously, which saves time, Dr. Levenson says. And “you can identify cells that co-express nuclear ER and PR without resorting to fluorescence-based methods, for example.”

“Using conventional color imaging, you would have to stain for ER and PR stains on successive sections, and would not be able to correlate co-expression on a cell-by-cell basis, since almost none of the cells in the first section will be included on the adjacent serial section,” Dr. Levenson says. So “all you could say is that the tumor is double-positive. Also, if you are seeking rare double-positive cells, you can’t expect to discover those with single stains on serial sections.”

Northwestern’s Dr. Kaplan says, “People think it’s important to identify whether an individual cell” rather than a tumor is positive for ER/PR in terms of developing “directed therapies going forward.”

In an application of digital pathology that may be out in five to 10 years, Dr. Kaplan is working with groups in Maryland and Texas to create “spectral signatures” for abnormal tissue that can “classify or qualify diagnoses more specifically.”

You use multispectral light to “shine through tissue and with a detector measure the spectral signature of pattern generated from that tissue section,” Dr. Kaplan explains. Most diagnoses—even a group of diagnoses in the same category—have their own spectral signature. The next step is to develop “libraries that match up with a specific diagnosis.”

The technology wouldn’t be useful in cases that are clearly benign or malignant, he says. But it could help pathologists navigate the more subjective areas of diagnosis. One prime example is dysplasia in Barrett’s esophagus, in which long-term gastric reflux leads to “intestinal metaplasia of the esophagus and that epithelium may become dysplastic or neoplastic,” Dr. Kaplan says.

Grading of the dysplasia in Barrett’s is subjective. “It’s generally recommended that a diagnosis of dysplasia have intramural group consensus and extramural review, if necessary,” he notes.

“The problem is that the [digital] data sets for endoscopic biopsies are very large files on the order of terabytes as opposed to gigabytes,” Dr. Kaplan says. But a single endoscopic biopsy could be subject to spectral analysis to see if any dysplasia is present—and, if so, the grading and whether there’s invasive carcinoma could be determined.

Other applications for multispectral imaging might include analyzing in situ ductal carcinoma with microinvasion or well-differentiated tumors where the differential includes benign processes, Dr. Kaplan says.

While initially he and his research colleagues are validating spectral signatures for a particular tumor or disease process by using stained tissues, they would eventually like to try it with unstained tissues.

“Anything that’s been done in any kind of digital imaging has involved some physical or chemical reaction: an H&E stain, fluorescent stain, or IHC stain. We’d like to eliminate the staining and create a digital signature with unstained tissue sections. There are two or three published papers on that,” he says.

Pathologists and industry experts agree on one central point about digital pathology: It isn’t and won’t become a replacement for pathologists who run the show when it comes to examining tissue to provide diagnoses and prognoses.

Worries that virtual microscopy will replace pathology are, in the view of Clarient’s Dr. Bloom, similar to the movie industry’s fear about DVDs. “They didn’t have to worry,” he says. “DVDs have [made] the desire to see movies even stronger and created a new revenue stream to boot.”