mug shot Joseph Y. Lo, Ph.D.

Assistant Professor,
Dept. of Radiology,
Duke University Medical Center

Assistant Professor,
Dept. of Biomedical Engineering,
Duke University School of Engineering

Director,
Biomedical Imaging and Medical Physics Third Year Study Program,
Duke University Medical School

Research Interests:
3D tomosynthesis imaging of the breast, computer-aided diagnosis, image processing

Contact Information:
Rm 161G Bryan Research Building
Box 3302 DUMC
Tel: (919) 684-PROF (7763)
Fax: (919) 684-7122
Email

updated Dec 4, 2005

Research Summary:

Check out the newly re-designed Duke Advanced Imaging Laboratories (DAILabs) website!

We are a multi-disciplinary team drawn from the radiology department of Duke University Medical Center and biomedical engineering department of Duke University School of Engineering. This program includes research in a wide range of medical physics topics, including digital chest and breast imaging.

Mammography is the modality of choice for early detection of breast cancer. In recent years, however, its limitations have been well publicized. About 1 in 5 breast cancers are missed during screening. There is an excessive rate where women are called back from screening for further diagnostic studies. Finally, as many as 4 out of 5 breast biopsies of suspicious lesions turn out to be benign and were thus arguably unnecessary.

We have been working to improve breast imaging in 2 key ways:

  1. 3D Tomosynthesis Imaging of the Breast
  2. Computer-aided Detection/Diagnosis

3D Tomosynthesis Imaging of the Breast

Breast tomosynthesis (or "tomo" for short) is one of the most exciting research developments in radiology in recent years. We acquire x-ray images taken from many different angles, then use that data to reconstruct 3D images of the breast. Breast tomo lets radiologists detect and characterize suspicious lesions better, because it removes overlapping normal tissue which might otherwise obscure the lesions. The goal is to provide 3D information at the same high resolution and reasonable dose as mammography, while possibly reducing compression for improved patient comfort. Since the system will be based on digital mammography, it will also be faster and cheaper than alternatives requiring dedicated equipment such as CT or MR. For these reasons, breast tomo may be the first technique that can actually replace mammography in the near future, providing improved sensitivity and specificity of breast cancer diagnosis.

I lead an interdisciplinary team consisting of many of the faculty and students of DAILabs, including Jim Dobbins, Ehsan Samei, and Carey Floyd. We are collaborating with Siemens Medical Solutions, a major industrial partner providing us with prototype hardware and invaluable scientific assistance. The photo on the right shows the system with the x-ray tube at the end of its 50 degree scan arc (+25 degrees from middle position).

Note to patients: This is an investigational prototype system, which means it is still being tested in research studies and is not yet approved by the U.S. Food and Drug Administration (FDA).

Here are some preliminary images from our on-going studies. This subject presented with a very subtle, indistinct mass as shown in the standard mammogram (left). Even with the magnification view, the mass is still very subtle (middle). The tomo scan easily reveals a spiculated mass which was later biopsied to reveal invasive ductal carcinoma.

Computer-aided Detection/Diagnosis

As mentioned above, although mammography is very sensitive at finding cancer, it results in many false positives. As few as 20% of currently biopsied cases actually reveal cancer. The remainder are all benign cases which underwent a potentially unnecessary surgical procedure. Preventing benign biopsies is the most important way to improve the efficacy of mammographic screening, especially as screening becomes more widespread.

We are developing machine learning and statistical models (such as artificial neural networks) to provide accurate diagnosis while being completely noninvasive. The models utilize existing, available information such as mammographic findings and patient history data. The computer models can provide radiologists, surgeons, and patients with information which was previously available only through biopsy, thus substantially reducing the number of unnecessary surgical procedures. These models are being developed using data from over 4,000 patients from Duke as well as other collaborating institutions. This is the largest multi-institution database of its type and represents the culmination of over 10 years of data collection. Here is a press release from the RSNA on a multi-institution evaluation of our predictive models which we published in 2002.

We have extended this work to ultrasound, which is used as an adjunct to mammography. Its advantages include the use of non-ionizing radiation, relatively low cost, and wide availability. We are building predictive models to identify probably benign breast masses using ultrasound findings. Some of our work was described in this WebMD article.

In addition, we are also studying image processing techniques to extract mammographic features such as masses and microcalcification clusters in an automated manner. These findings may serve as inputs to the diagnostic models above, or act as an independent "second reader" for the early detection of breast cancer.

Finally, we continue to apply our experience in artificial intelligence and image processing to a wide range of other clinically significant problems, including the detection of lung nodules in chest radiography, outcome analysis for patients with acute pancreatitis based upon CT and clinical findings, and optimized treatment planning for radiation oncology.

Our work has been supported by many external grants from the National Institutes of Health / National Cancer Institute, the US Army Breast Cancer Research Program, the Susan G. Komen Breast Cancer Foundation, the Whitaker Foundation for biomedical engineering research, and other sources.

Sample Publications

These are recent papers selected to cover a representative range of topics. For more information, see Dr. Lo's CV for a full list of about 30 papers which may be found in any large medical library, as well as a list of Dr. Lo's grant funding.

Resources and Environment

The Duke Advanced Imaging Labs (DAILabs) is a group of over 20 faculty, staff and students devoted to full-time research. It is part of the new Duke Medical Phyiscs Program. Duke Hospital, one of the largest private hospitals in the United States, is part of Duke University Health System and currently is licensed for over 1000 beds. The division of breast imaging in the Department of Radiology performs over 20,000 exams each year, approximately 500 of which are sent to biopsy each year. DAILabs occupies over 2,000 square feet of laboratory and office space in two buildings. Computer facilities include a variety of workstations which are connected through a high speed dedicated network to all of the radiology department's clinical image acquisition systems, such as digital chest radiography, magnetic resonance imaging, ultrasound, nuclear medicine, and computed tomography. This network also includes radiographic film digitizers and printers, and a state of the art image archiving and retrieval system.

Check out the newly re-designed Duke Advanced Imaging Laboratories (DAILabs) website!

[ work | CV | publications | personal ]