A new imaging method could one day help some people who’ve had treatment for rectal cancer avoid unnecessary surgeries after complete tumor destruction from chemoradiation.
The new method can differentiate between rectal tissues with residual cancers and those without tumors after chemotherapy and radiation.
Rectal cancer, along with colon cancer, is the third-most common type of cancer in the United States, and treatment and surgery greatly affect patients’ quality of life.
Quing Zhu, professor of biomedical engineering at the Washington University in St. Louis McKelvey School of Engineering, and members of her lab developed a system using a new imaging technique—acoustic resolution photoacoustic microscopy co-registered with ultrasound (AR-PAM/US) and paired with a “deep learning” artificial intelligence neural network.
This technique was better able to determine the presence of residual tumors in treated rectal tumor bed tissues than other types of imaging, such as MRI, which is often unable to discern residual cancer from scar tissue.
Results of the research—the first feasibility study using AR-PAM imaging in patients with rectal cancer previously treated with radiation and chemotherapy—appear in the journal Radiology.
“Our PAM/US system paired with the deep learning neural network has great potential to better identify patients suitable for nonoperative management and improve patient quality of life,” Zhu says. “If we can tell after radiation and chemotherapy which patients may have a good response with no residual tumors, the patient may be able to avoid surgery.”
The team spent more than three years investigating this technology in surgically removed colon and rectum specimens with promising results before developing the prototype for patient studies, says William Chapman Jr., a resident physician in surgery and one of several School of Medicine clinicians at Barnes-Jewish Hospital in St. Louis to help conduct the yearlong prospective study in patients with rectal cancer.
“We hope that improved imaging provided by AR/PAM will significantly improve our ability to discriminate between patients with residual tumors and those who have been completely cured without surgery,” he says. “By both avoiding morbid, unneeded surgery and reducing the burden of surveillance testing, photoacoustic imaging could be a leap forward in the current management of locally invasive rectal cancer.”
In the study, after completing chemotherapy and radiation, patients underwent PAM/US imaging with a handheld endorectal laser probe developed in Zhu’s lab. The probe has a rotating head that allows for a 360-degree image of the rectum, the last 6 inches of the colon. The end of the probe, which takes one image per second, is covered by a small latex balloon inflated with water that allows transmission of the ultrasound and photoacoustic waves to the rectal wall. These waves highlight changes in the vasculature in the tissue and as well as new tumor growth. The imaging procedure added about 20 minutes that patients were under anesthesia.
Xiandong Leng, co-first author and doctoral student, was instrumental in system and software development, Zhu says. He designed and built the AR/PAM endoscope—the first of its kind—and programmed the system to acquire data and process and display images in real time.
“From the very preliminary ex vivo data, my setup clearly disclosed multi-layer structure from ultrasound image and rich blood vessels in the submucosa of normal colorectal tissue,” Leng says. “In contrast to normal tissue, malignant tumor bed shows a lack of multilayer structure and blood vessels. This important finding may reveal an important feature accessing patients’ treatment response to chemotherapy and radiation therapy.”
In the first phase of the study, the team used data from surgically removed tissue specimens from more than 2,000 images from 22 patients to train the neural network, an artificial intelligence-based set of algorithms that operates similar to the human brain, to recognize normal and cancerous colorectal tissue. In the second phase, they used images from the living tissue from 10 patients who had previously undergone chemotherapy and radiation. Several hundred images from five of those patients were used to fine-tune the neural network, and hundreds of images from five patents were withheld for testing.
The deep-learning PAM model, which co-first author Shihab Uddin designed and developed, correctly predicted the cancerous status of all five of the patients who had undergone imaging, while the MRI images misclassified three out of five patients, and the ultrasound-only deep-learning model incorrectly declared three patients as cancer free.
Mutch says the team is very optimistic about the results.
“This is spectacular news, and it moves us closer in the transition from concept to clinically useful technology,” he says. “The hope is that it will allow us to differentiate patients who had a complete response to chemotherapy and radiation from those patients with residual tumor. This will help better determine which patients can be managed nonoperatively versus those who truly need an operation.”
Going forward, the team plans to conduct a clinical study to confirm these initial results in a large group of rectal cancer patients who have completed chemotherapy and radiation and will undergo surgery or follow patients after treatment.
Funding for this research came from the Siteman Cancer Center and the Foundation for Barnes-Jewish Hospital, as well as the National Cancer Institute.
