An array microscope for ultrarapid virtual slide processing and telepathology. Design, fabrication, and validation study

This paper describes the design and fabrication of a novel array microscope for the first ultrarapid virtual slide processor (DMetrix DX-40 digital slide scanner). The array microscope optics consists of a stack of three 80-element 10 x 8-lenslet arrays, constituting a "lenslet array ensemble." The lenslet array ensemble is positioned over a glass slide. Uniquely shaped lenses in each of the lenslet arrays, arranged perpendicular to the glass slide constitute a single "miniaturized microscope." A high-pixel-density image sensor is attached to the top of the lenslet array ensemble. In operation, the lenslet array ensemble is transported by a motorized mechanism relative to the long axis of a glass slide. Each of the 80 miniaturized microscopes has a lateral field of view of 250 microns. The microscopes of each row of the array are offset from the microscopes in other rows. Scanning a glass slide with the array microscope produces seamless two-dimensional image data of the entire slide, that is, a virtual slide. The optical system has a numerical aperture of N.A.= 0.65, scans slides at a rate of 3 mm per second, and accrues up to 3,000 images per second from each of the 80 miniaturized microscopes. In the ultrarapid virtual slide processing cycle, the time for image acquisition takes 58 seconds for a 2.25 cm2 tissue section. An automatic slide loader enables the scanner to process up to 40 slides per hour without operator intervention. Slide scanning and image processing are done concurrently so that post-scan processing is eliminated. A virtual slide can be viewed over the Internet immediately after the scanning is complete. A validation study compared the diagnostic accuracy of pathologist case readers using array microscopy (with images viewed as virtual slides) and conventional light microscopy. Four senior pathologists diagnosed 30 breast surgical pathology cases each using both imaging modes, but on separate occasions. Of 120 case reads by array microscopy, there were 3 incorrect diagnoses, all of which were made on difficult cases with equivocal diagnoses by light microscopy. There was a strong correlation between array microscopy vs. "truth" diagnoses based on surgical pathology reports. The kappa statistic for the array microscopy vs. truth was 0.96, which is highly significant (z=10.33, p <0.001). There was no statistically significant difference between rates of agreement with truth between array microscopy and light microscopy (z=0.134, p >0.05). Array microscopy and light microscopy did not differ significantly with respect to the number/percent of correct decisions rendered (t=0.552, p=0.6376) or equivocal decisions rendered (t=2.449, p=0.0917). Pathologists rated 95.8% of array microscopy virtual slide images as good or excellent. None were rated as poor. The mean viewing time for a DMetrix virtual slide was 1.16 minutes. The DMetrix virtual slide processor has been found to reduce the virtual slide processing cycle more than 10 fold, as compared with other virtual slide systems reported to date. The virtual slide images are of high quality and suitable for diagnostic pathology, second opinions, expert opinions, clinical trials, education, and research.