{"forum": "7EPItQwtu1", "submission_url": "https://openreview.net/forum?id=t4CzelLd5X", "submission_content": {"authorids": ["yil8@uci.edu", "wenwu.ye@jfhealthcare.com", "jin.yao@jfhealthcare.com", "hui.xue@jfhealthcare.com"], "abstract": "Localizing thoracic diseases on chest X-ray plays a critical role in clinical practices such as diagnosis and treatment planning. However, current deep learning based approaches often require strong supervision, e.g. annotated bounding boxes, for training such systems, which is infeasible to harvest in large-scale. We present Probabilistic Class Activation Map (PCAM) pooling, a novel global pooling operation for lesion localization with only image-level supervision. PCAM pooling explicitly leverages the excellent localization ability of CAM (Zhou et al., 2016) during training in a probabilistic fashion. Experiments on the ChestX-ray14 (Wang et al., 2017) dataset show our method outperforms state-of-the-art baseline on the localization task. Visual examination on the probability maps generated by PCAM pooling shows clear and sharp boundaries around lesion regions compared to the localization heatmaps generated by CAM.\n", "paper_type": "methodological development", "TL;DR": "We present Probabilistic-CAM (PCAM) pooling, a new global pooling operation to explicitly leverage the localization ability of CAM during training, and show that PCAM pooling improves weakly supervised lesion localization on ChestX-ray14 dataset.", "authors": ["Yi Li", "Wenwu Ye", "Jin Yao", "Hui Xue"], "track": "short paper", "keywords": ["Chest X-rays", "Lesion localization", "Weakly supervised learning"], "title": "Weakly Supervised Lesion Localization With Probabilistic-CAM Pooling", "pdf": "/pdf/c37c0b33d7ee3e9770b88a41d2e83f9363ad30c4.pdf", "paperhash": "li|weakly_supervised_lesion_localization_with_probabilisticcam_pooling", "_bibtex": "@misc{\nli2020weakly,\ntitle={Weakly Supervised Lesion Localization With Probabilistic-{\\{}CAM{\\}} Pooling},\nauthor={Yi Li and Wenwu Ye and Jin Yao and Hui Xue},\nyear={2020},\nurl={https://openreview.net/forum?id=t4CzelLd5X}\n}"}, "submission_cdate": 1579955667102, "submission_tcdate": 1579955667102, "submission_tmdate": 1587172143175, "submission_ddate": null, "review_id": ["JgMMnRr3Xc", "SPYZh2Tckv", "46yeDB7p4h", "_PS-oesvHX"], "review_url": ["https://openreview.net/forum?id=t4CzelLd5X&noteId=JgMMnRr3Xc", "https://openreview.net/forum?id=t4CzelLd5X&noteId=SPYZh2Tckv", "https://openreview.net/forum?id=t4CzelLd5X&noteId=46yeDB7p4h", "https://openreview.net/forum?id=t4CzelLd5X&noteId=_PS-oesvHX"], "review_cdate": [1584341972861, 1584163615226, 1584025338803, 1583785844381], "review_tcdate": [1584341972861, 1584163615226, 1584025338803, 1583785844381], "review_tmdate": [1585229867890, 1585229867386, 1585229866894, 1585229866403], "review_readers": [["everyone"], ["everyone"], ["everyone"], ["everyone"]], "review_writers": [["MIDL.io/2020/Conference/Paper91/AnonReviewer3"], ["MIDL.io/2020/Conference/Paper91/AnonReviewer2"], ["MIDL.io/2020/Conference/Paper91/AnonReviewer1"], ["MIDL.io/2020/Conference/Paper91/AnonReviewer4"]], "review_reply_count": [{"replyCount": 0}, {"replyCount": 0}, {"replyCount": 0}, {"replyCount": 0}], "review_replyto": ["7EPItQwtu1", "7EPItQwtu1", "7EPItQwtu1", "7EPItQwtu1"], "review_content": [{"title": "Potentially interesting; but lack of implementation details and the experimental results are not significant", "review": "It is potentially interesting to normalize logits and make the CAM more interpretable. \nHowever, it is not clear how to implement the proposed method. To be specific, the weights for the pooling layer and the network outputs depend on each other (illustrated as a loop in Figure 1). I am not sure how to train this network in an end-to-end fashion. It would be more clear to provide more details in the caption of Figure 1.\nAdditionally, the experimental results are not promising. Based on Table 1 and Figure 2, the P-CAM has higher accuracy compared to the baseline but has poorer performance on the false positive rate. Ergo, the significance of P-CAM is not fully justified.\n", "rating": "2: Weak reject", "confidence": "4: The reviewer is confident but not absolutely certain that the evaluation is correct"}, {"title": "Weakly supervised lesion localization on chest x-ray", "review": "The key idea is to add a loop in the model to utilize the cam as a weight to improve the localization capability of weakly supervised method. The method is applied to ChestX-ray14 dataset and compared with simple CAM. The method shows potential better localization capability in a weakly supervised fashion.\n\nThe method in principle is able to produce sharper \u2018heatmaps\u2019 for lesion localization than simple CAM. However, the false positives seem pretty severe, just looking at Figure 2. I am not sure which one is more preferred especially if the threshold can be adjusted. I am also concerned about the impact on classification accuracy with the introduced structure. Classification accuracy is also important but not reported in the paper.\n\nDetailed comments:\n- How was the 0.9 threshold determined? This seems to be an important (hyper)parameter. And what\u2019s the threshold for the regular CAM? \n- More weakly supervised methods in medical imaging should be compared or discussed.\n- Figure 1 is confusing. Maybe try to color code to differentiate the classification and CAM paths. \n- Curious to see how the results compare with simply squaring weights in the original CAM.\n\n", "rating": "2: Weak reject", "confidence": "5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature"}, {"title": "Simple and effective", "review": "Prop:\n1. This paper propose a probabilistic-CAM pooling to bridge the pixel-level localization and image-level classification. Normalized weights are used for weighted average pooling. PCAM explicitly leverages the CAM in a probabilistic fashion.\n\nCons:\n1. One advantage of PCAM is that it is trained like a probabilistic model. However, no significant improvement are shown that PCAM is accurate than other CAMs. Larger output leads to better IoBB and worse false positive.\n2. I think the author should explicitly point out the difference between this work and related work(Ilse  et  al.,  2018). Why sigmoid  function is used for bounding. Tanh and sigmoid are used in (Ilse  et  al.,  2018). \n3. I'm wondering whether PCAM can contribute to the classification accuracy compared to other CAMs.\n\nOther comments:\n1. Cannot understand \"This may explain the fact that PCAM pooling has relatively larger average false positives than CAM with LSE pooling\", this means PCAM is worse than LSE pooling ?\n2. More detailed description of the architecture is preferred. The architecture seems to be different with the architecture used in (Ilse  et  al.,  2018). More details would help reproduce the work.\n", "rating": "2: Weak reject", "confidence": "3: The reviewer is fairly confident that the evaluation is correct"}, {"title": "A not so new CAM pooling", "review": "This paper presents a pooling strategy for Class Activation Maps (CAM) to learn to localize thoracic diseases using image-level supervision. The evaluation is performed using the ChestX-ray14 dataset.\n\nOverall, the paper is clear and easy to read but misses on a large part of the literature in weakly supervised learning.\n\nMy main criticisms are:\n1. The idea is not novel and should not be presented as such. Several papers have been published using similar or identical poolings in both audio and image processing.\n2. The method is compared against only one result (Wang et al., 2017) which was obtained on the ChextX-ray8 dataset, while the paper is mentioning ChestX-ray14.\n3. The AFP is much higher than the LSE pooling. It is not surprising for a method with much higher AFP to also have higher IoBB. Generally, the results are not convincing.\n\nOther remarks:\n1. (Ilse et al., 2018) should not be the citation for the MIL framework.\n2. The MIL framework is not necessarily assigning attention weights to each embedding. This is one way to do it but there are others in the literature.\n3. Several choices that were made for the evaluation are not justified: the choice of 0.9 for the threshold, the comparison of IoBB > 0.5 (a table in supplementary material could have been added for other values).", "rating": "2: Weak reject", "confidence": "5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature"}], "comment_id": [], "comment_cdate": [], "comment_tcdate": [], "comment_tmdate": [], "comment_readers": [], "comment_writers": [], "comment_reply_content": [], "comment_content": [], "comment_replyto": [], "comment_url": [], "meta_review_cdate": 1585631277577, "meta_review_tcdate": 1585631277577, "meta_review_tmdate": 1585631277577, "meta_review_ddate ": null, "meta_review_title": "MetaReview of Paper91 by AreaChair1", "meta_review_metareview": "All four reviewers recommend 'weak reject', citing weakness in the methodological novelty and experimental sufficiency. The AC concurs with the reviewing opinons.", "meta_review_readers": ["everyone"], "meta_review_writers": ["MIDL.io/2020/Conference/Program_Chairs", "MIDL.io/2020/Conference/Paper91/Area_Chairs"], "meta_review_reply_count": {"replyCount": 0}, "meta_review_url": ["https://openreview.net/forum?id=t4CzelLd5X&noteId=yLp3cs82_6K"], "decision": "reject"}