- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1(−1)
Read review
Paper summary
EC-MIL introduces a MIL framework that separates the estimation of instance-class-specific contribution from the context of other instances. This framework provides the possibility of intrinsically explaining the signed contribution of each instance to the bag prediction in a single forward pass and is adaptable to different MIL backbone architectures.
Critical strengths
- Targeting instance-level explanation in MIL architectures as a critical requirement for medical and biomedical fields.
- The proposed method is flexible to be adjusted to different backbone architectures (ABMIL, TransMIL, and etc) and not limited to a specific backbone design.
- Providing an efficient single forward pass solution for instance-level explanation during inference.
- Wide evaluation covering different backbone architectures, datasets, and components of the proposed method.
- The source code is provided.
Critical weaknesses
-
Insufficient comparison with competing intrinsic attribution methods (Additive MIL is insufficiently compared, and AttriMIL [1] is not included at all). In particular, there is no systematic comparison of the proposed approach with competing intrinsic methods under the same backbones and datasets. Therefore, the evaluation does not sufficiently establish the advantages of EC-MIL over its main intrinsic competitors.
-
In Section 4.2, Assumption C is strong and is not guaranteed by the proposed architecture. The authors aim to prove that a low risk of bag prediction implies low instance attribution error. Their proof assumes that the attribution error of each instance does not change too much when other instances change. But even if the evidence () remains fixed, its context weight () can change significantly (under context effect). Therefore, the attribution error may also change. The authors should explain why Assumption C is expected to hold for EC-MIL, or evaluate it on synthetic data where the true instance contributions are known.
-
The authors claim that WE is a signed attribution for each class, but their theoretical result in Section 4.2 concerns the difference between the contributions of two classes. If we add the same value to the evidence of an instance for both classes, the logit difference and therefore the cross-entropy loss will not change, while the WE of each class changes. Therefore, bag-level CE cannot guarantee the accuracy or sign of each class-specific WE.
-
The experimental results show that EC-MIL improves performance in some settings but leads to underperformance in others. The authors should provide an explanation of what limits or boosts the performance of their method in those cases. Since the proposed method aims to improve interpretability, understanding its own behavior and limitations is also important. For example, EC-MIL behaves differently on BRACS vs. CAMELYON16. The question is: which characteristics of these datasets relate to the positive or negative performance changes of the proposed model compared to the baseline backbone?
-
The writing is hard to follow. The introduction and related work need to be polished to cover only the relevant explanation. Also, in the introduction, the distinction between post-hoc and intrinsic approaches is not clarified. For example, in Section 2.2, the descriptions and limitations of GI and IG are combined into one long sentence with multiple citations, while they could be explained in two shorter sentences.
-
The complete performance comparisons should be moved to the main text, as the differences are difficult to see in the charts. Figure 2 has poor presentation. The differences are not visible. Consider magnifying the relevant performance range (e.g., above 50%) or showing the performance differences separately.
-
A statistical comparison would be beneficial to assess the significance of the improvements provided by the proposed method.
-
While WE is supposed to be signed contributions, Figure 4a shows only positive WE values. A distribution plot of the computed WE values, showing the proportion and magnitude of positive and negative contributions, would help clarify how signed attribution behaves in practice.
Reference: [1] AttriMIL: Revisiting attention-based multiple instance learning for whole-slide pathological image classification from a perspective of instance attributes, L. Cai et al., Medical Image Analysis 2025
Questions for authors
Please refer to weaknesses.
10 comments
Rebuttal Table 1: Prediction and attribution comparison of intrinsic methods across four synthetic tasks. Values for AdditiveMIL and EC-MIL on the first three tasks are from the original manuscript; AttriMIL and Pos-Neg-L results are newly added. Bold indicates the highest mean AUC and AUPRC-2 and the lowest mean AUPC for each task.
| Task | Model | AUC | AUPRC-2 | AUPC |
|---|---|---|---|---|
| 4-Bags | AdditiveMIL (ABMIL) | 0.9916 0.0015 | 0.9053 0.0370 | 0.3464 0.0169 |
| 4-Bags | AttriMIL | 0.9966 0.0006 | 0.7615 0.0284 | 0.6244 0.0842 |
| 4-Bags | EC-ABMIL | 0.9932 0.0018 | 0.9307 0.0074 | 0.3618 0.0164 |
| 4-Bags | EC-TransMIL | 0.9942 0.0011 | 0.9386 0.0066 | 0.3635 0.0058 |
| 4-Bags | EC-ILRAMIL | 0.9950 0.0014 | 0.9295 0.0172 | 0.3654 0.0040 |
| Pos-Neg | AdditiveMIL (ABMIL) | 0.9492 0.0063 | 0.8891 0.0145 | 0.2599 0.0190 |
| Pos-Neg | AttriMIL | 0.9679 0.0040 | 0.7869 0.0422 | 0.4148 0.0566 |
| Pos-Neg | EC-ABMIL | 0.9702 0.0059 | 0.9351 0.0052 | 0.2396 0.0067 |
| Pos-Neg | EC-TransMIL | 0.9652 0.0039 | 0.8998 0.0096 | 0.2641 0.0093 |
| Pos-Neg | EC-ILRAMIL | 0.9729 0.0054 | 0.9183 0.0079 | 0.2533 0.0107 |
| Adjacent-Pairs | AdditiveMIL (ABMIL) | 0.7803 0.0592 | 0.6027 0.0593 | 0.5767 0.0956 |
| Adjacent-Pairs | AttriMIL | 0.9456 0.0291 | 0.5751 0.0139 | 0.8404 0.0551 |
| Adjacent-Pairs | EC-ABMIL | 0.8149 0.0111 | 0.8000 0.0205 | 0.4689 0.0588 |
| Adjacent-Pairs | EC-TransMIL | 0.9878 0.0039 | 0.7970 0.0137 | 0.4470 0.0186 |
| Adjacent-Pairs | EC-ILRAMIL | 0.9910 0.0028 | 0.7967 0.0167 | 0.4360 0.0108 |
| Pos-Neg-L | AdditiveMIL (ABMIL) | 0.8660 0.0066 | 0.9403 0.0032 | 0.3251 0.0087 |
| Pos-Neg-L | AttriMIL | 0.8798 0.0058 | 0.8331 0.0464 | 0.4485 0.0337 |
| Pos-Neg-L | EC-ABMIL | 0.8777 0.0065 | 0.9824 0.0032 | 0.3471 0.0092 |
| Pos-Neg-L | EC-TransMIL | 0.9849 0.0023 | 0.9743 0.0043 | 0.3085 0.0076 |
| Pos-Neg-L | EC-ILRAMIL | 0.8939 0.0067 | 0.9773 0.0049 | 0.3514 0.0104 |
The additional results show that EC-ABMIL achieves higher AUPRC-2 than AdditiveMIL and AttriMIL on all four synthetic tasks. For example, on 4-Bags, AttriMIL achieves a classification AUC of 0.9966 and a native-attribution AUPRC-2 of 0.7615, compared with 0.9932 and 0.9307 for EC-ABMIL. On Pos-Neg-L, EC-ABMIL achieves an AUPRC-2 of 0.9824, compared with 0.9403 for AdditiveMIL and 0.8331 for AttriMIL. Jointly reporting these metrics evaluates predictive ability and instance evidence recovery separately.
Rebuttal Table 2: Prediction and deletion faithfulness of intrinsic methods on MUT. AdditiveMIL and EC-MIL values are from the original manuscript; AttriMIL is newly added. Bold indicates the highest mean AUC and the lowest mean AUPC for each feature setting.
| Feature | Model | AUC | AUPC |
|---|---|---|---|
| scGPT | AdditiveMIL (ABMIL) | 0.9442 0.0033 | 0.0190 0.0010 |
| scGPT | AttriMIL | 0.9123 0.0047 | 0.6027 0.0310 |
| scGPT | EC-ABMIL | 0.9482 0.0026 | 0.0232 0.0004 |
| scGPT | EC-TransMIL | 0.9620 0.0040 | 0.0257 0.0006 |
| scGPT | EC-ILRAMIL | 0.9752 0.0025 | 0.0322 0.0010 |
| text2vec | AdditiveMIL (ABMIL) | 0.9500 0.0022 | 0.0193 0.0007 |
| text2vec | AttriMIL | 0.9185 0.0030 | 0.6156 0.0379 |
| text2vec | EC-ABMIL | 0.9501 0.0048 | 0.0230 0.0007 |
| text2vec | EC-TransMIL | 0.9626 0.0026 | 0.0257 0.0005 |
| text2vec | EC-ILRAMIL | 0.9748 0.0022 | 0.0308 0.0019 |
The MUT results further show the prediction and attribution performance of the intrinsic methods. With scGPT and text2vec, EC-ABMIL achieves classification AUCs of 0.9482 and 0.9501, compared with 0.9123 and 0.9185 for AttriMIL. Their corresponding native-attribution AUPCs are 0.0232 and 0.0230 for EC-ABMIL, and 0.6027 and 0.6156 for AttriMIL. AdditiveMIL achieves lower AUPCs of 0.0190 and 0.0193 in these two settings. These deletion results are included in revised Figure 5, with complete numerical results in revised Appendix Table A15.
We also include AdditiveMIL and AttriMIL in the new TCGA-LUAD TP53 mutation prediction task, extending intrinsic-method comparisons to another real-world WSI setting. This task predicts molecular status from WSIs using the ResNet18 and pathology-pretrained DINO (SSL) features used in the original manuscript, with frozen feature extractors and the same data splits. Prediction and native-attribution deletion results are reported in Rebuttal Table 3, extending the comparisons to molecular prediction from pathology images. In the revised manuscript, Table 2 reports backbone-averaged classification performance for the original and EC variants, Figure 7 shows deletion faithfulness, and Appendix J Tables A20–A21 provide the complete per-model results.
Rebuttal Table 3: Comparison of intrinsic methods on TCGA-LUAD TP53 mutation prediction.
| Feature | Model | AUC | Intrinsic Attribution AUPC |
|---|---|---|---|
| ResNet18 | AdditiveMIL (ABMIL) | 0.6403 0.0353 | 0.3781 0.0728 |
| ResNet18 | AttriMIL | 0.6404 0.0298 | 0.7922 0.0667 |
| ResNet18 | EC-ABMIL | 0.5867 0.0418 | 0.2629 0.2559 |
| ResNet18 | EC-TransMIL | 0.5572 0.0239 | 0.3286 0.1643 |
| ResNet18 | EC-ILRAMIL | 0.6325 0.0666 | 0.3602 0.1851 |
| DINO (SSL) | AdditiveMIL (ABMIL) | 0.6883 0.0559 | 0.3055 0.1131 |
| DINO (SSL) | AttriMIL | 0.7307 0.0312 | 0.8016 0.0932 |
| DINO (SSL) | EC-ABMIL | 0.7635 0.0272 | 0.3213 0.1122 |
| DINO (SSL) | EC-TransMIL | 0.7267 0.0393 | 0.3341 0.1191 |
| DINO (SSL) | EC-ILRAMIL | 0.7459 0.0253 | 0.4511 0.0595 |
With ResNet18 features, AdditiveMIL and AttriMIL achieve higher classification AUCs than the three EC variants, while WE achieves lower AUPCs (0.2629, 0.3286, and 0.3602) than their native attributions (0.3781 and 0.7922). With DINO features, all three EC variants achieve higher AUCs than AdditiveMIL, and EC-ABMIL and EC-ILRAMIL also exceed AttriMIL. AdditiveMIL has the lowest deletion AUPC of 0.3055; WE ranges from 0.3213 to 0.4511, all below AttriMIL at 0.8016. These results extend the intrinsic-method comparison to molecular prediction with two pathology feature representations, jointly reporting prediction and attribution performance.
W2: What Assumption C Constrains, How It Relates to EC-MIL, and Empirical Validation
(1) Our theory aims to explain why instance contributions can be learned using only bag labels. A single bag prediction constrains the sum of contributions, allowing instance-level errors to cancel each other out. EC-MIL shares local evidence across bags, so the same instance information is repeatedly supervised by bag labels under different companions. Its contribution decomposition is therefore jointly constrained across combinations. Assumption C characterizes a regime where contextual error cancellation is controlled, linking excess bag-level cross-entropy risk to instance decomposition error. This insight provides a theoretical explanation of how bag supervision can train meaningful instance contributions.
Regarding your point about changes in context weights, the true contributions themselves can also change with companions. Learning this variation is precisely the model's objective. When predicted contributions track changes in the true contributions, weights can change substantially while contribution errors remain small. Assumption C constrains the relative variation of this tracking error across contexts and is therefore compatible with the Context Stream substantially adjusting weights according to instance relations.
Thank you for recognizing the importance of instance-level explanations in MIL, the flexibility of EC-MIL across backbones, and its efficient single-forward-pass attribution. Below, we address your concerns and questions point by point.
W1: Systematic Comparison with Competing Intrinsic Attribution Methods (AdditiveMIL and AttriMIL)
(1) Our original manuscript already compares EC-MIL with AdditiveMIL using the same ABMIL backbone across prediction, evidence recovery, and deletion evaluation. Table 1 in the original manuscript reports classification AUC and attribution AUPRC-2 on three synthetic tasks, with EC-ABMIL exceeding AdditiveMIL on both metrics in all three tasks. For example, Adjacent-Pairs AUC is 0.8149 vs. 0.7803, and AUPRC-2 is 0.8000 vs. 0.6027. Original Appendix Tables A19–A25 also report AdditiveMIL prediction or attribution results on synthetic tasks, MUT, and CAMELYON16 with ResNet18 features. The original manuscript thus includes intrinsic-method comparisons using the same backbone, as well as prediction and explanation evaluations across tasks.
(2) Thank you for suggesting the inclusion of AttriMIL and a more systematic comparison. In this rebuttal, we include AttriMIL and compare it jointly with AdditiveMIL and EC-MIL using the same tasks, input features, and data splits. We use the ABMIL implementation of AdditiveMIL and EC-ABMIL for comparison within the same backbone, and also report AttriMIL, EC-TransMIL, and EC-ILRAMIL to show the prediction and attribution performance of the intrinsic methods. AdditiveMIL uses its native Patch-Logits, AttriMIL uses its native instance attribution, and EC-MIL uses WE. The table below reports classification AUC, recovery of known instance contributions (AUPRC-2), and deletion AUPC. Revised Table 1 and Figure 3 include AttriMIL and all four synthetic tasks; complete classification, evidence-recovery, and deletion results with standard deviations are consolidated in revised Appendix Tables A8 and A11–A12.
The synthetic evaluation covers the three original tasks and adds Pos-Neg-L. Digits 1, 2, and 3 each contribute +1; digits 4, 5, and 6 each contribute −1; and digits 0, 7, 8, and 9 contribute 0. When digits 1, 2, and 3 form an L shape with 2 at the corner, each participating instance contributes +2. The bag label is determined by the sum of instance contributions. This evaluates local instance evidence and three-instance spatial relations within the same task. On MUT, we use the scGPT and text2vec features from the original manuscript to further compare prediction and deletion faithfulness on a real-world task.