Fig 1.
Block diagram of the proposed deep vector-based convolutional neural network for classification of induced pluripotent stem cell colonies.
Fig 2.
Quantitative feature measurements for a healthy image of a segmented induced pluripotent stem cell colony.
(A) Original image. (B) Iterative thresholding. (C) Morphological operation with size filter (D) Labeling.
Fig 3.
Quantitative feature measurements for an unhealthy image of a segmented induced pluripotent stem cell colony.
(A) Original image. (B) Iterative thresholding. (C) Morphological operation with size filter (D) Labeling.
Fig 4.
Fisher scores assigned to each feature in the feature sets of (A) morphology, and (B) textures.
Cen, centroid; Are, area; Ecc, eccentricity; Per, perimeter; Ori, orientation; Maj, major axis; Min, minor axis; Dia, equivalent diameter; Sol, solidity; Ext, extent; D_V, difference variance; Hom, homogeneity; Ene, energy; D_E, difference entropy; Con, contrast; Cor, correlation; Inf_1, information measure of correlation_1; S_A, sum average; Inf_2, information measure of correlation_2; S_E, sum entropy; Ent, entropy; S_V, sum variance; Var, variance.
Fig 5.
V-CNN architecture for recognition of induced pluripotent stem cell colony quality.
Table 1.
Performance validation of each individual feature, using the area under the curve (AUC), corresponding standard error (SE), and 95% confidence interval.
Fig 6.
Comparison of the ranges of values between healthy and unhealthy colony groups based on (A) morphological and (B) textural features.
Fig 7.
Receiver operating characteristic curve for the morphological features.
Fig 8.
Receiver operating characteristic curve for the textural features.
Fig 9.
Performance of the V-CNN model with regard to accuracy and loss.
The model’s performance in classifying the colonies was based on (A) morphological, (B) textural, and (C) combined features.
Table 2.
Performance of the proposed V-CNN model and SVM classifier in classifying colonies on the basis of morphological, textural, and combined features.
Table 3.
Five-fold cross-validation of the performance of the proposed V-CNN model and SVM classifier in classifying colonies on the basis of their morphological, textural, and combined features.