We propose a high-throughput 3D imaging cytometer for fast quantification of DNA double strand break (DSB) frequency in cells for DNA damage study. With structured illumination enabled depth contrast and a fast focus tunable lens enabled scanning, this system generates a three-dimensional stack of clustered nuclei γH2AX foci with submicron resolution at a speed of 800 cells/second. Moreover, we unify the stack construction with the deep neural network, which largely improve quantification accuracy as well as the processing speed. Compared to previous 2D imaging approach, the addition of z-resolution in our 3D method provides an extra dimension of contrast and thus allows for more accurate DNA DSB quantification.
Ionising radiation causes various types of DNA damages including double strand breaks (DSBs). DSBs are often recognized by DNA repair protein ATM which forms gamma-H2AX foci at the site of the DSBs that can be visualized using immunohistochemistry. However most of such experiments are of low throughput in terms of imaging and image analysis techniques. Most of the studies still use manual counting or classification. Hence they are limited to counting a low number of foci per cell (5 foci per nucleus) as the quantification process is extremely labour intensive. Therefore we have developed a high throughput instrumentation and computational pipeline specialized for gamma-H2AX foci quantification. A population of cells with highly clustered foci inside nuclei were imaged, in 3D with submicron resolution, using an in-house developed high throughput image cytometer. Imaging speeds as high as 800 cells/second in 3D were achieved by using HiLo wide-field depth resolved imaging and a remote z-scanning technique. Then the number of foci per cell nucleus were quantified using a 3D extended maxima transform based algorithm. Our results suggests that while most of the other 2D imaging and manual quantification studies can count only up to about 5 foci per nucleus our method is capable of counting more than 100. Moreover we show that 3D analysis is significantly superior compared to the 2D techniques.