Figure 1.
Schematic illustration of the experimental scheme for the prediction of osteogenic differentiation potential using multiple and time-course morphological features.
hBMSCs were cultured in non-induction medium in first 4 days, then the medium was replaced with osteogenic induction medium only for the Induction sample. From day 0 to day 14, cell images were automatically acquired by BioStation CT every 8 hours. ALP activity and calcium deposition rates were evaluated on days 14 and 21, respectively. Using multiple morphological features covering 2 weeks culture, two types of hBMSC osteogenic differentiation evaluation results were predicted by individual prediction models.
Figure 2.
Experimentally determined biological results after the osteogenic differentiation.
A: Experimentally determined ALP activity rate on day 14 of differentiation. B: Experimentally determined calcium deposition rate on day 21 of differentiation.
Figure 3.
Phase contrast raw image from BioStation CT and its processed image.
The images of beginning (day 1), middle (day 3 and 7), and the end (day 13) in the induction period of Lot 1 are indicated as examples. Raw images were binarized with MetaMorph.
Figure 4.
Time series changes of characteristic morphological features.
From the 9 morphological features measured, elliptical form factor (A) and fiber breadth (B) of Lot 1 are indicated as representative examples. The symbols indicate the mean value of each morphological feature from all cells in one condition (3 wells ×5 view fields). Roughly, 4,000 to 40,000 cells were measured for the mean. Standard deviations are shown as error bars.
Figure 5.
Schematic illustration of two scenarios examined to simulate clinical feasibility.
A: (Scenario I) New patient prediction scheme: Trained by historical patient dataset only. Images from all passages of patient 3 were used for prediction. B: (Scenario II) Ongoing patient prediction scheme: Trained by historical patient datasets and a partial dataset from the new patient. For example, for the prediction of cell potential of patient 3, Scheme I uses images of patient 1 and 2 only. Scheme II used images of patient 1 and 2, together with some images from patient 3.
Figure 6.
Prediction accuracies in the new patient scheme.
A: Scatter plot of experimentally determined values versus predicted values in D14_ALP model, B: Scatter plot of experimentally determined values versus predicted values in D21_Ca model.
Figure 7.
Detailed prediction results in new patient scheme.
A: Prediction results and error range in the D14_ALP model. B: Prediction results and error range in the D21_Ca model. All the plotted data were rearranged in the order of experimental values.
Table 1.
Prediction accuracy of Ridge regression models for osteogenic differentiation status of hBMSCs.
Figure 8.
Prediction accuracies in the ongoing patient scheme.
A: Scatter plot of experimentally determined values versus predicted values in D14_ALP model, B: Scatter plot of experimentally determined values versus predicted values in D21_Ca model.
Figure 9.
Detailed prediction results in ongoing patient scheme.
A: prediction results and error range in the D14_ALP model. B: Prediction results and error range in the D21_Ca model. All the plotted data were rearranged in the order of experimental values.
Table 2.
Prediction accuracy of Redge regression models with elimination of each individual features for ALP activity rate.
Table 3.
Prediction accuracy of Redge regression models with elimination of each individual features for calcium deposition rate.
Figure 10.
A: Schematic illustration of cell image processing.
The raw images were first pre-processed by background reduction processing by deconvolution and open-close filters. Then, images were binarized by the optimized threshold. The noisy objects were eliminated by particle deletion filter. B: Schematic illustration of cell morphology measurements and data processing. In all object recognized images, all existing objects were measured for the 9 morphological features. Since 1 condition was designed to consist of 3 wells ×5 view fields, all the corresponding object measurement results were processed as a same sample. The average and standard deviation within one sample of all morphological features at each time point were used as the input features for modeling. C: Schematic illustration of prediction model construction. Prediction of differentiation potential consisted of two steps. First, two types of prediction models (D14_ALP model or D21_Ca model) were constructed with the set of image data and experimental evaluation. Second, the values of D14_ALP or the D21_Ca were predicted from the input features of the sample targeted for prediction. The predicted biological rates are compared to the experimentally-determined results to evaluate the accuracy of prediction model.