DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Status
Claims 1-30 in the claim set filed May 20th, 2025, were pending for examination in the Application No. 17/398,655 filed August 10, 2021. In the remarks and amendments received on February 20th, 2026, claims 1-30 are amended. Accordingly, claims 1-30 are currently pending for examination in the application.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on March 5th, 2026, is in compliance with the provisions of 37 CFR 1.97. Accordingly, the IDS is being considered and attached by the examiner.
Response to Amendment
Applicant’s amendments filed February 20th, 2026, to the Claims have overcome each and every objection previously set forth in the Non-Final Office Action mailed August 20th, 2025. Accordingly, the objection(s) are withdrawn in response to the remarks and amendments filed. Examiner warmly thanks Applicant for considering the suggested amendments to be made to the disclosure.
Response to Arguments
Applicant’s arguments filed February 20th, 2026, regarding the rejection(s) of the independent claims have been fully considered but are moot because the arguments do not apply to the new combination of the references—i.e., further in view of Shao et al. (Shao; “TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classification,” 2021; provided by Applicants’ IDS filed April 25, 2023)—being used in the current rejection below.
Claim Objections
Claims 5 and 16 are objected to because of the following informalities failing to comply with 37 CFR 1.71(a) for "full, clear, concise, and exact terms" (see MPEP § 608.01(m)):
In claim 5, the phrase “the pseudo-labels” in the claim should be “the one or more pseudo-labels” to maintain consistency in language within the claims; and
In claim 16, the phrase “the patches” in the claim should be “the one or more patches” to maintain consistency in language within the claims.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-30 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (Sharma; “Cluster-to-Conquer: A Framework for End-to-End Multi-Instance Learning for Whole Slide Image Classification,” 2021) in view of Alzubaidi et al. (Alzubaidi; “Review of deep learning: concepts, CNN architectures, challenges, applications, future directions,” 2021), and further in view of Shao et al. (Shao; “TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classification,” 2021; provided by Applicants’ IDS filed April 25, 2023).
Regarding claim 1, Sharma discloses one or more processors, comprising:
circuitry to:
partition one or more images into a plurality of patches (1st para. of pg. 4, recite(s)
[1st para. of pg. 4] “…Given a WSI
W
(bag) with label
y
, we extract
w
1
,
w
2
,
w
3
,
…
,
w
n
patches (instances) from it for training. As we approach the classification problem with the MIL framework, positive bags include at least one diseased patch (instance) while negative bags contain all healthy patches (instances).”
, where extracting “patches (instances)” from a work slide image (“WSI”) is partitioning one or more images into a plurality of patches);
identify, using object identification, one or more objects depicted by one or more patches of the plurality of patches (1st para. of pg. 4—see citation in preceding limitation immediately above—, where section 3.1 on pgs. 3-4 further recites:
[3.1. Problem Background] “For digital pathology classification problems, WSIs (W) of patients are available along with their disease labels. …patches containing substantial tissue area…”
, where the “classification” of image patches of “tissue” of a patient as either “healthy” or “diseased” is identifying one or more objects (e.g., “tissue”) using object identification (e.g., “classification”));
predict, for the one or more patches, one or more object classes for the one or more identified objects (1st para. of pg. 4—see citation in limitation “partition one or more images…” above—, where the classification of patches into “healthy” or “diseased” patches is predicting one or more object classes for the one or more identified objects (e.g., tissue));
generate one or more (1st para. of pg. 4—see citation in limitation “partition one or more images…” above—, where determining the labels of “positive” or “negative” are labels generated for one or more of the plurality of patches based on the one or more predicted object classes (e.g., “healthy” or “diseased”)); and
update(line 5 of pg. 4 and section 3.5 on pg. 6, recite(s)
[line 5 of pg. 4] “To this end, we have developed a convolutional neural network framework, C2C,…”
[section 3.5 on pg. 6] “…The model was implemented with PyTorch and trained on a single RTX2080 GPU. The framework was trained end-to-end with Adam optimizer with a batch size of 1 and a learning rate of
1
e
-
4
for 30 epochs. Empirically,
α
=
1
,
β
=
0.01
and
γ
=
0.1
for loss hyperparameters demonstrated best performance. We experimented with different k. …”
, where the “convolutional neural network framework, C2C,” is one or more neural networks and “train[ing]” the framework using a “learning rate” is updating the one or more neural networks) based, at least in part, on:
first loss indicating an accuracy of the object identification technique within the one or more patches (1st para. of pg. 6, recite(s)
[1st para. of pg. 6] “…Along with WSI and patch cross-entropy loss, for each cluster, KL-divergence loss between the patches attention weight and a uniform distribution is included. The inclusion of KL-divergence loss regularizes the same cluster patches attention distribution and allows the attention module to weight all the positive class patches uniformly. The aggregated loss can be written as:
L
G
y
,
G
y
'
,
G
a
,
G
e
=
α
*
L
W
S
I
+
β
*
L
P
a
t
c
h
+
γ
*
L
K
L
D
where
α
,
β
, and
γ
balance the importance of different losses.”
, where the “patch-level cross-entropy loss” (e.g.,
L
P
a
t
c
h
) is a first loss indicating an accuracy of object identification (e.g., classification) within one or more patches”); and
second loss indicating an accuracy of an enumeration of instances of the one or more objects within the one or more images based, at least in part, on one or more (1st para. of pg. 6—see citation in preceding limitation immediately above—, where the “WSI… cross-entropy loss…” (e.g., “
L
W
S
I
”) is a second loss indicating an accuracy of an enumeration of instances of one or more objects within the one or more images (e.g., tissue) based on at least labels (e.g., “weighting” based on the one or more labels of “positive” and “negative”) of the one or more objects and one or more weight factors (e.g., “attention weight[s]”) associated with the one or more labels as further recited in the 1st para. of pg. 8:
[1st para. of pg. 8] “…We observed that C2C could accurately identify the patches with tumor regions and assign them higher attention weights.”
).
Where Sharma does not specifically disclose
…to update one or more parameters of one or more neural networks;
Alzubaidi teaches in the same field of endeavor of training one or more neural networks
…to update one or more parameters of one or more neural networks (section “Optimizer selection” on pg. 22, recite(s)
[section “Optimizer selection”] “…Loss functions, which are founded on numerous learnable parameters (e.g. biases, weights, etc.) or minimizing the error (variation between actual and predicted output), are the core purpose of all supervised learning algorithms. The techniques of gradient-based learning for a CNN network appear as the usual selection. The network parameters should always update though all training epochs, while the network should also look for the locally optimized answer in all training epochs in order to minimize the error.
The learning rate is defined as the step size of the parameter updating. The training epoch represents a complete repetition of the parameter update that involves the complete training dataset at one time. Note that it needs to select the learning rate wisely so that it does not influence the learning process imperfectly, although it is a hyper-parameter…”
, where “parameter updating” is updating one or more parameters of a neural network).
Since Sharma also discloses training the one or more neural networks using an optimizer comprising of a learning rate and training epochs, a person of ordinary skill in the art before the effective filing date of the claimed invention would have recognized that the training of the one or more neural networks of Sharma comprises of updating one or more parameters of the one or more neural networks as taught by Alzubaidi above.
Where Sharma in view of Alzubaidi does not specifically disclose
generate one or more pseudo-labels for one or more of the plurality of patches based on the one or more predicted object classes;
Shao teaches in the same field of endeavor of multiple instance learning (MIL)
generate one or more pseudo-labels for one or more of the plurality of patches based on the one or more predicted object classes (section 2.1 on pg. 2 of Shao, recite(s)
[section 2.1 on pg. 2] “The application of MIL in WSIs can be divided into two categories. The first one is instance-level algorithms [7, 23, 24, 25, 26], where a CNN is first trained by assigning each instance a pseudo-label based on the bag-level, and then the top-k instances are selected for aggregation. …”
, where “each instance” is a patch).
Since Sharma also discloses using the multiple instance learning (MIL) framework to approach the classification problem (1st para. of pg. 4—see citation in limitation “partition one or more images…” above), a person of ordinary skill in the art before the effective filing date of the claimed invention would have recognized that labels for the patches into positive and negative bags using the MIL framework as disclosed in the system of Sharma in view of Alzubaidi are pseudo-labels based on the bag-level (e.g., WSI) label as taught by Shao above.
Regarding claim 2, Sharma, as modified by Alzubaidi and Shao, discloses the one or more processors of claim 1, wherein Sharma further discloses the one or more neural networks are trained to account for dependencies between the one or more patches of the one or more images (1st para. of pg. 4—see citation in claim 1 limitation “partition one or more images…” above—, where the “MIL” (Multiple Instance Learning) framework for training is training the one or more neural networks to account for dependencies (e.g., positive and negative) between the one or more patches of the one or more images).
Regarding claim 3, Sharma, as modified by Alzubaidi and Shao, discloses the one or more processors of claim 2, wherein Sharma further discloses the one or more neural networks include one or more embedded(section 3.3. on pg. 4, recite(s)
[section 3.3. on pg. 4] “A patch-level encoder,
G
e
x
;
Θ
e
:
x
→
h
, maps all patches to
l
-dimensional embeddings where
Θ
e
is the set of trainable parameters. …”
, where the “patch-level encoder” is one or more embedded encoder blocks to encode the dependencies between the patches (e.g., “maps all patches to
l
-ddimensional embeddings”)).
Where Sharma, as modified by Alzubaidi and Shao, does not specifically disclose
…embedded transformer encoder blocks…;
Shao further teaches in the same field of endeavor of multiple instance learning (MIL)
…embedded transformer encoder blocks… (section 3.2 on pg. 5, recite(s)
PNG
media_image1.png
411
1220
media_image1.png
Greyscale
).
Since Sharma also discloses using the MIL framework to approach the classification problem (1st para. of pg. 4—see citation in claim 2 above), it would have been obvious to one of ordinary skill in the art before the effective filing date of the presently filed invention to further modify the system of Sharma, as modified by Alzubaidi and Sharma, to substitute the patch-level encoder with one or more embedded transformer blocks to improve using the MIL framework by adding positional information to further increase the use of sequential order information when encoding the dependencies between patches while still yielding the predictable result of encoding the dependencies between the patches.
Regarding claim 4, Sharma, as modified by Alzubaidi and Shao, discloses the one or more processors of claim 1, wherein the one or more neural networks are to determine the one or more pseudo-labels for the one or more patches (1st para. of pg. 4—see citation in claim 1 limitation “partition one or more images…” above—and line 5 of pg. 4 and section 3.5 on pg. 6—see citation in claim 1 limitation “update one or more parameters…” above—of Sharma discloses determining labels (e.g., “positive” and “negative”) for the one or more patches (e.g., “instances”) by one or more neural networks (e.g., a “convolutional neural network framework, C2C,”); wherein section 2.1 on pg. 2 of Shao teaches the labels as “pseudo-labels” as taught in claim 1 above—see teaching of Shao in the combination of Sharma in view of Alzubaidi and Shao above in claim 1 above).
Regarding claim 5, Sharma, as modified by Alzubaidi and Shao, discloses the one or more processors of claim 4, wherein Shao further teaches the circuitry is to further train the one or more neural networks using image-level labels for the one or more images and the one or more pseudo-labels for the one or more patches (subheading “Problem Formulation” in section 3.1 “Correlated Multiple Instance Learning” on pg. 3, recite(s)
PNG
media_image2.png
431
1196
media_image2.png
Greyscale
, where the problem of “The instance-level labels…are unknown, and the bag level label” is known (i.e., one of the “binary MIL classification”) is training the one or more neural networks using image-level labels (i.e., “bag-level label[s]”) for the one or more images (i.e., WSIs) and the pseudo-labels (i.e., predicted “instance-level” labels) for the patches (i.e., instances)).
Regarding claim 6, Sharma, as modified by Alzubaidi and Shao, discloses the one or more processors of claim 4, wherein Sharma further discloses the one or more neural networks are trained using instance-wise loss supervision based, at least in part, upon the one or more pseudo-labels (1st para. of pg. 6—see citation in claim 1 above—, where
[section 3.4 on pg. 5] “…WSI prediction probability and the instance representation are passed through
G
y
'
:
z
→
y
'
to obtain patches prediction probability. Instance loss is included with weak supervision assumption. …”
, where the “patch cross-entropy loss” or “instance loss” is instance-wise loss supervision based, at least in part, upon the pseudo-labels (e.g., the positive and negative labels of the patches recited in the 1st para. of pg. 4—see citation in claim 1 limitation “partition one or more images…” above)).
Regarding claim 7, the claim recites similar limitations to claim 1 but in the form of a system comprising the one or more processors of claim 1. Therefore, claim 7 recites similar limitations to claim 1 and is rejected for similar rationale and reasoning (see the analysis for claim 1 above).
Regarding claim 13, the claim is the method performed by the one or more processors of claim 1. Therefore, claim 13 recites similar limitations to claim 1 and is rejected for similar rationale and reasoning (see the analysis for claim 1 above).
Regarding claim 19, the claim recites similar limitations to claim 1 but in the form of a non-transitory machine-readable medium. Therefore, claim 19 recites similar limitations to claim 1 and is rejected for similar rationale and reasoning (see the analysis for claim 1 above).
Regarding claim 25, the claim recites similar limitations to claim 1 but in the form of an image classification system further comprising a memory for storing the one or more parameters of the one or more neural networks for the one or more neural networks. Alzubaidi further teaches said memory (3rd para. of pg. 58 and section “GPU-based approach” on pg. 59, recite(s)
[3rd para. of pg. 58] “In addition to the computational load cost, the memory bandwidth and capacity have a significant effect on the entire training performance, and to a lesser extent, deduction. More specifically, the parameters are distributed through every layer of the input data, there is a sizeable amount of reused data, and the computation of several network layers exhibits an excessive computation-to-bandwidth ratio. By contrast, there are no distributed parameters, the amount of reused data is extremely small, and the additional FC layers have an extremely small computation-to-bandwidth ratio. Table 3 presents a comparison between different aspects related to the devices. In addition, the table is established to facilitate familiarity with the tradeoffs by obtaining the optimal approach for configuring a system based on either FPGA, GPU, or CPU devices. It should be noted that each has corresponding weaknesses and strengths; accordingly, there are no clear one-size-fts-all solutions.”
[section “GPU-based approach”] “GPUs are extremely effective for several basic DL primitives, which include greatly parallel-computing operations such as activation functions, matrix multiplication, and convolutions [326–330]. Incorporating HBM-stacked memory into the up-to-date GPU models significantly enhances the bandwidth. Tis enhancement allows numerous primitives to efficiently utilize all computational resources of the available GPUs. Te improvement in GPU performance over CPU performance is usually 10-20:1 related to dense linear algebra operations.”
). Since Sharma discloses implementing and training the model on a GPU, a person of ordinary skill in the art before the effective filing date of the claimed invention would have recognized that the system of Sharma further comprises a memory for storing the one or more parameters of the one or more neural networks as taught by Alzubaidi above.
Therefore, claim 25 recites similar limitations to claim 1 and is rejected for similar rationale and reasoning (see the analysis for claim 1 above).
Regarding claims 8, 14, 20, and 26, the claims recite similar limitations to claim 2 and are rejected for similar rationale and reasoning (see the analysis for claim 2 above).
Regarding claims 9, 15, 21, and 27, the claim recites similar limitations to claim 3 and is rejected for similar rationale and reasoning (see the analysis for claim 3 above).
Regarding claims 10, 16, 22, and 28, the claim recites similar limitations to claim 4 and is rejected for similar rationale and reasoning (see the analysis for claim 4 above).
Regarding claims 11, 17, 23, and 29, the claim recites similar limitations to claim 5 and is rejected for similar rationale and reasoning (see the analysis for claim 5 above).
Regarding claims 12, 18, 24, and 30, the claim recites similar limitations to claim 6 and is rejected for similar rationale and reasoning (see the analysis for claim 6 above).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JULIA Z YAO whose telephone number is (571)272-2870. The examiner can normally be reached Monday - Friday (8:30AM - 5PM).
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Emily Terrell can be reached on (571)270-3717. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/J.Z.Y./Examiner, Art Unit 2666
/MING Y HON/Primary Examiner, Art Unit 2666