Fig 1.
Our proposed approach showing the role of the cuboid nominator in selecting cuboids from the input CT scan, while also extracting meaningful volume-level representation.
The selected cuboids are then fed into a cuboid classifier, which extracts cuboid-level features. The volume-level representation is then concatenated with the sum of the cuboid-level representation and their corresponding positional embedding to provide the patient-level classification.
Fig 2.
An illustration of the infection clusters impact from two cuboids cf and cl on cuboid cn along the z axis (Left).
The relationship between δ and ε for a given probability x = 0.8 along with the impact of cf and cl (Right).
Table 1.
Classification results on the following two settings: (1) Three combined infection severity classes from the the HUST dataset [7]: control, Type I (Mild and medium cases), and Type II (Severe and critical cases). (2) The same three classes, but from the combined HUST and MosMed datasets. †Based on the confusion matrix provided by the authors in the supplementary materials of [7]. ‡Based on 5 runs with different random seeds to overcome the non-deterministic 3D convolutional.
Fig 3.
Confusion matrices for our method on the (a) HUST [7] and (b) MosMed [27] datasets.
Fig 4.
Explainability visualization as a result of highlighting the contents of the selected cuboids (red) compared to manually annotated infection mask (green).
(a) shows two validation samples with the highest infection coverage score while (b) shows the two validation samples with the lowest infection coverage score.
Fig 5.
The impact of varying the k number of nominated cuboids on: (a) the training and validation infection coverage rate during the training process.
(b) The validation F1 measure, with the size of the dots indicating how early in the training process, the best F1 measure was achieved.
Table 2.
Classification results on the following three settings: (1) Three combined infection severity classes from the the HUST dataset [7]: control, Type I (Mild and medium cases), and Type II (Severe and critical cases). (2) The same three classes, but from the combined HUST and MosMed datasets. (3) All five classes, control, mild, medium, severe and critical, from the same split of the combined HUST and MosMed datasets.
Fig 6.
(a) The impact of changing the epoch at which the proximity factor for infection clusters is activated on the validation infection coverage and F1 measure.
(b) The ablation results when removing the influence of the increase and decrease coefficients from (2) on the validation infection coverage as well as the impact of removing the volume-level features. (c) Comparison between the validation infection coverage pattern during training of the three and five class settings.
Fig 7.
The confusion matrix with mean and standard deviations for the five (a) and three (b) classes in the HUST, MosMed dataset combination when testing our proposed approach using 5-fold cross validation.