Researchers at University of Illinois at Urbana–Champaign and at Chicago have tested a new optical method for diagnosing breast cancer that could lead to an automated process for rapidly and more accurately determining whether breast-tissue lesions are benign or malignant.
In a recent special section of the Journal of Biomedical Optics, Hassaan Majeed, a PhD candidate in bioengineering at Urbana–Champaign, and colleagues describe their use of quantitative optical-phase measurements for cancer diagnosis.
When an abnormality is discovered during a screening procedure such as a mammography, standard practice is for the physician to take a tissue biopsy, which is then stained using hematoxylin and eosin (H&E). The staining provides enough contrast for a pathologist to study key morphological features of the tissue under a microscope.
The tissue analysis is done manually; because of variations in things like staining intensity and the illumination used, the process does not lend itself to automation. Manual inspections are also subject to investigator bias, and the process is time-consuming. This can, in some cases, result in late disease diagnosis — a critical shortcoming given that early diagnosis is said to significantly improve chances of survival.
In “Breast cancer diagnosis using spatial light interference microscopy,” researchers used the breast-tissue biopsies of 400 different patients, selecting two parallel, adjacent sections from each biopsy. One of these sections was stained using H&E, leaving the other unstained.
The unstained samples (bottom row in figure below) were analyzed using a spatial light interference microscopy (SLIM) module attached to a commercial phase-contrast microscope. The system generated different interferograms for each tissue sample using shifting patterns of scattered and unscattered light. Four interferograms, corresponding to four different phase shifts, were used to produce one quantitative-phase image.
On this SLIM image, areas with different refractive properties appear in different colors. The boundary between the tumor and its extracellular environment is clearly delineated, making it possible to assess whether the tumor is malignant.
As part of the experiment, two board-certified pathologists were trained to interpret the tissue morphological details from SLIM images by being shown H&E and SLIM images of the same tissue cores, labeled benign or malignant, side by side.
At the testing stage, each pathologist was first shown the stack of SLIM images for 109 selected tissue cores. The pathologists classified each core as either benign or malignant. Then the process was repeated for the stack of H&E images for the same 109 cores. Using each pathologist’s diagnosis of the H&E stained cores as the gold standard, the researchers measured the success of diagnosis using SLIM images by counting the number of agreements between SLIM- and H&E-based diagnoses.
The agreement between SLIM- and H&E-based diagnosis was 88% for the first pathologist and 87% for the second. Agreement between the two pathologists when rating SLIM images stood at 83%, whereas the same for H&E images was much higher at 98%. The authors of the paper expect the agreement between the diagnoses of the two pathologists on SLIM images would increase significantly if they received more training in interpreting SLIM images.
Because this diagnosis method is based on quantitative data, rather than on a pathologist making a subjective assessment, the researchers believe that these preliminary results show the potential of their technique to become the basis for an automated image-analysis system that would provide a fast and accurate diagnosis.
Coauthors are SPIE Fellow Gabriel Popescu, Mikhail Eugene Kandel, Kevin Han, Zelun Luo, Virgilia Macias, Krishnarao Tangella, and Andre Balla.