Fig 1.
Experimental set-up for high-throughput monitoring of spheroid formation.
The system is composed of three major components: (i) a microscopic set-up equipped with a motorized stage and an incubator chamber. (ii) a well plate containing agarose microwell inserts for spheroid culturing. (iii) a computer with image acquisition and analysis software. In the bottom-right corner, an overview of the culture conditions applied in this study is shown. For each condition, the corresponding well number of the 24 well plate is indicated.
Fig 2.
General overview of the proposed segmentation approach.
(A) Raw image. (B) Gradient image computed from the raw image using Roberts gradient operator [50]. The gradient output was not normalised. (C) Circular mask of detected microwell. For further operations, the radius of the circular mask was set to the inner radius of the microwell (i.e. 75 pixels using the described microscopic set-up and microwells with a diameter of 200 μm). (D) The microwell mask was used to extract the intensity values of the corresponding spheroid and its surroundings (i.e. the background inside the microwell). Based on these intensity values, an intensity threshold was automatically computed using Otsu’s method [51]. This adaptive threshold was used to segment the raw image. (E) Evaluating the noise of the background, the gradient image was segmented with a fixed threshold of 0.020. (F) Both segmented images (D, E) were combined using an OR-operator. (G) The spheroid of interest (the one corresponding to the circular mask) was extracted using an AND-operator. (H) The final spheroid mask was obtained by applying the following morphological operations: (i) The largest object was retained together with all the objects that have an area smaller or equal to 25 pixels. (ii) Unconnected pixels were bridged and the resulting mask was diagonally filled to eliminate the 8-connectivity of the background. (iii) Holes smaller or equal to 150 pixels were filled, objects were morphologically eroded with a disk of radius 2 pixels, the largest object was retained and then morphologically dilated with a disk of radius 2 pixels. (iv) Prior to filling up the remaining holes, the mask was bridged and diagonally filled for a second time.
Fig 3.
Qualitative assessment of the automatic segmentation approach.
Semi-transparent, red-coloured masks of segmented spheroids overlaid on their corresponding raw image for: (A) Set 1; initial, small-sized spheroids. (B) Set 2; final, small-sized spheroids. (C) Set 3; initial, large-sized spheroids. (D) Set 4; final, large-sized spheroids. The black scale bars represent 200 μm.
Table 1.
Relative error values obtained for the extracted features.
Table 2.
Sensitivity (TPR) and precision (PPV) of the automatic segmentation approach.
Table 3.
Comparative study between manual and automatic segmentation according to their time consumption.
Fig 4.
The average dynamic response of the spheroid area and circularity for each condition.
(A), (B) Average response of the area for the two agarose types, respectively for the small- and large-sized spheroids. (C), (D) Average response of the circularity for the two agarose types, respectively for the small- and large-sized spheroids. (E), (F) Images of the cell aggregation process over time for the two agarose types, respectively for the small- and large-sized spheroids. The black scale bars represent 200 μm.
Fig 5.
Illustration of the off-line monitoring process of individual spheroids in function of time.
During cell aggregation, the dynamic response of the spheroid area and circularity was monitored. Here, six large-sized spheroids cultured on agarose type B were examined over time. Screenshots were taken after (A) 6 and (B) 12 hours. The black scale bars represent 200 μm.