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 .
Claims 1-20 are currently pending in U.S. Patent Application No. 18/918,636 and an Office action on the merits follows.
Drawings
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(4) because reference character “512” has been used to designate both “USE ADJUSTED SEGMENTATION MODEL TO GENEATE SET OF RESULTS” (Fig. 3) and “MODIFY THE INITIAL SEGMENTATION MODEL USING THE SET OF CONTOURED STRUTURES VIA A TRANSFER LEARNING TECHNIQUE TO OUTPUT A PERSONLIZED SEGMENTATION MODEL FOR THE USER” (Fig. 5). Examiner also notes that Applicant’s Specification at [0039] discloses “Optionally, at 312, the process 300 can involve using the resulting adjusted segmentation model to generate a set of results from the validation subset to be used to compare against the accuracy and specified minimum performance claims consistent with a specified clearance of the model.” Accordingly, Examiner understands those later two labels of Fig. 3 (‘510’ and ‘512’), to potentially/likely be mislabeled, but at the minimum either Fig. 3 label 512 should be corrected/replaced (with ‘312’), and/or the text description should be modified to match that of 512 if 512 is the more accurate/appropriate label. Applicant’s Specification at [0038] also discloses label/identifier ‘310’ that, similar to 312, is not currently reflected in the Figures/Fig. 3 (further suggesting Applicant may intend for Fig. 3 510 and 512 to read 310 and 312 respectively). Examiner cautions against simply replacing the 510 label of Fig. 3 to read 310, without addressing/verifying the associated text description, because that change would provoke a similar objection since 37 CFR 1.84(p)(4) reads “(4) The same part of an invention appearing in more than one view of the drawing must always be designated by the same reference character, and the same reference character must never be used to designate different parts.” To summarize, three changes are understood to be appropriate, 1) change identifier/label 512 to read 312 in Fig. 3 (text description can remain as-is in view of [0039]), 2) change the identifier/label 510 in Fig. 3 to read 310, and 3) modify the text description (Fig. 3) accompanying 310.
Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 10-18 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter, specifically “at least one machine-readable medium” comprising transitory signal embodiments. See MPEP § 2106.03. In the context of the flowchart illustrated in MPEP 2106, analysis at Step 1 of the Subject Matter Eligibility test follows the ‘No’ path (Step 1: No). Applicant’s Specification at e.g. [0055], [0057], etc., disclose non-transitory medium embodiments that constitute statutory subject matter, however these examples are non-limiting, there is no express disavowal and/or redefinition of the term “machine-readable medium” to include only those statutory embodiments, and the broadest reasonable interpretation of the claim in light of the specification (MPEP 2173.01) concludes that the claim as a whole covers transitory signal(s)/media, which do not fall within the definition of a process, machine, manufacture or composition of matter (In re Nuijten, 500 F.3d 1346, 84 USPQ2d 1495 (Fed. Cir. 2007)).
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim(s) 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim(s) 1/10/19, at line 7 recites the limitation in part “a structure model”. It is not clear what constitutes a “structure model”, even and particularly in view of a plain meaning reading, absent an ordinary and customary meaning known to and/or readily recognized by those of ordinary skill in the art (‘ordinary’ and ‘customary’). With reference to MPEP 2111.01 sub-section V, illustrating that summary flow used in ascertaining proper claim interpretation, analysis follows that annotated version produced below. See also MPEP 2173.02 sub-section I, identifying how claims under examination are construed differently and with a potentially lower threshold for ambiguity.
During prosecution the Office construes claims by giving them their broadest reasonable interpretation consistent with the specification in an effort to establish a clear record of what the applicant intends to claim. Such claim construction during prosecution may effectively result in a lower threshold for ambiguity than a court's determination. Packard, 751 F.3d at 1323-24, 110 USPQ2d at 1796-97 (Plager, J., concurring). However, applicant has the ability to amend the claims during prosecution to ensure that the meaning of the language is clear and definite prior to issuance or provide a persuasive explanation (with evidence as necessary) that a person of ordinary skill in the art would not consider the claim language unclear. In re Buszard, 504 F.3d 1364, 1366, 84 USPQ2d 1749, 1750 (Fed. Cir. 2007)
An internet-based search of the term in question suggests that it does not have any ‘ordinary and customary’ meaning in the art, nor is it a term familiar to, and/or frequently encountered by the Examiner (without any assertions regarding how the Examiner’s skill in the art may compare to that of a hypothetical POSITA). The language in question is indefinite at least because the specification does not clearly set forth a definition for the term(s), and the record as a whole does not lend clarity to what does and does not constitute a ‘structure model’ (e.g. does it require some shape/volume decoder portion?) sufficient to "inform those skilled in the art about the scope of the invention with reasonable certainty" Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 899, 110 USPQ2d at 1689 (2014). Limited support present appears in e.g. [0026], for which model 114 is alternatively identified as a “2D to 3D model”. Examiner additionally notes for clarity of record purposes, that while Applicant may intend for ‘a set of contoured structures’ to be 3D in nature, permissible interpretation at least for the case of claim(s) 1/10/18 arguably includes 2D structures.
Dependent claims 2-9, 11-18 and 20 inherit and fail to cure that/those deficiencies identified above for the case of claim(s) 1/10/19 and are rejected accordingly. Claim(s) 9/18 further limit functions associated with the ‘structure model’ (114) in question, but only in terms of an associated function and not so as to sufficiently resolve that ambiguity identified above, concerning what such a model necessarily is or itself necessarily requires.
PNG
media_image1.png
830
908
media_image1.png
Greyscale
Claim Rejections - 35 USC § 103
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 of this title, 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.
1. Claims 1-8, 10-17 and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Fuchs et al. (US 2021/0295528 A1) in view of Alzate-Grisales et al. “SAM-UNETR: Clinically significant prostate cancer segmentation using transfer learning from large model”.
As to claim 1, Fuchs discloses a method comprising:
sending, to a user, an initial segmentation model, trained using a first set of planning image structures (Fig. 7, initial segmentation model at 704, trained on that initial/first training dataset at 702, said dataset at 702 further comprising a ‘first’ set of structures of interest within a domain (and useable in a treatment planning more broadly) – e.g. the ‘first set’ may be any of those structures of interest/indicative of risk (e.g. associated with a ‘tumor’ annotation in Fuchs as compared to ‘non-tumor’), [0048] “Deep Interactive Learning (DIaL) may be used to minimize pathologists' annotation time by iteratively annotating mislabeled regions to improve a model. DIaL may be used with a pretrained model from a different cancer type to reduce manual training annotation on pancreatic pathology images”), and 702 further comprising a ‘second set’ of structures, e.g. those not indicative of risk, non-tumor, healthy/normal, or alternatively/more broadly any of a second class of interest (ideal interpretation for the ‘second set’ is likely those structures associated with e.g. any non-preferred model output, e.g. those structures (e.g. challenging or rare) warranting user correction when supplying feedback data of 708). [0052] Initial Training disclosure, [0054], [0125] “The computing system may use the training dataset to train the image segmentation model. For each sample image of the training dataset, the computing system may apply the image segmentation model to generate a segmented image ( e.g., the segmented image 518'). The segmented image may have an area ( e.g., the first area 524A or the second area 524B) determined to correspond to one of the regions of interest and have another area (e.g., the third area 524C) determined to not correspond to any of the regions of interest in the sample image. The computing system may compare the areas of the segmented image with the annotations for the corresponding sample image to determine a loss metric. Using the loss metric, the computing system may update at least one of the parameters of the image segmentation model”, etc.,; Examiner notes regarding that ‘sending’ see Fig. 7 706, Fig. 8 etc.,; Examiner asserts housing the model/copies either client side, and/or centrally/remotely from the client/user, are both included in permissible interpretation, provided the user/client has access to the pre/initially trained model/model results (e.g. arguing the model is not itself sent, may raise questions regarding as to if the model must then not exist at any location other than in the possession of the user to which it was sent));
receiving, from the user, adjusted WSI patch/sub-images, the adjusted patch images being modified by the user (Fig. 7 708 receive feedback dataset, wherein the feedback dataset has been ‘adjusted’ to further include the user-provided annotation feedback (additional forms of ‘adjustment’/deformation/data augmentation, etc., may also read, see Fuchs at [0028]), [0054] “Correction Characteristic features are annotated during initial annotation, but challenging or rare features may not be included. During the correction step, these challenging features that the model could not predict correctly are annotated to be included in the training set to improve the model. In this step, the annotators look at segmentation predictions and correct any mislabeled regions. If the predictions are satisfied throughout training images, the model is finalized”, [0125-0127], etc.,);
generating, using the adjusted selected images as an input to a feedback handler, new annotations (Fig. 7 712 following that decision to retrain 710, wherein feedback handler 508 in conjunction with feedback interface 514 serve to provide feedback dataset 530 to model trainer 506, [0080] “For example as depicted, the first new annotation 522'A may be associated with a first condition corresponding to the first region of interest 520A. Furthermore, the second new annotation 522'B may be associated with a second condition corresponding to the second region of interest 520B. The new annotations 522' may be included in the feedback dataset 530”, etc.,); and
modifying, using processing circuitry, the initial segmentation model using the feedback dataset via a transfer learning technique to output a personalized segmentation model for the user (Fig. 5B, Fig. 7, re-iterated Training at 704 following and based on those newly identified annotations 712, wherein the ‘personalization’ is that the model, after subsequent training, better/more accurately segments those ‘challenging or rare’ features of interest to the user(s) supplying the feedback (Applicant’s Specification makes reference to a ‘segmentation style’ that may be less expressly captured in the ‘generating’ step above – while also disclosing the user supplied feedback allows the user to embed, into the fine-tuned model, ‘specific knowledge and expertise’ [0019]); Transfer Learning specific disclosure in Fuchs appears in at least e.g. [0048], [0061] “In one test run, the cohort contained 759 cases with pancreatic ductal adenocarcinomas whose primary sites are pancreas. 14 whole slide images for training, and 23 whole slide images for numerical evaluation. A pretrained breast model was fine-tuned using DIaL to segment pancreatic carcinomas. During the first iteration, a pathologist annotated false positives on non-tumor subtypes that are not presented on breast training images. The first correction took an hour (example depicted in FIG. 2B). During the second iteration, the pathologist annotated false negatives on pancreatic carcinomas. The second correction took two hours. The pathologist spent total 3 hours to annotate 14 pancreatic pathology images”; Examiner further notes that the claims as interpreted under BRI may not limit which domains are involved (cross-organ, cross-imaging modality, etc.,), and/or any specifically identified TL technique itself, however Fuchs discloses at least transfer between breast and pancreatic domains).
Fuchs further fails to explicitly disclose (in view of struck limitations as presented above) generating any ‘set of contoured structures’ that are 3D in nature (not expressly/explicitly recited), in addition to selected slices from a volume.
Alzate-Grisales evidences the obvious nature of transfer learning as applied in a segmentation model context (page 118218 “Against this backdrop, our research is centered in two main goals: • Exploiting transfer learning in high-end architectures: Large companies have invested heavily in developing and training complex DL models that, while incredibly powerful, are often beyond the reach of many researchers due to their computational requirements. This study seeks to leverage these efforts through the use of transfer learning. By leveraging the pre-trained weights of these sophisticated models, we aim to initiate and adapt them for PCa imaging tasks, bypassing the enormous computational overhead typically associated with training such models from scratch … ” page 118219 “Building on this potential, we present a DL-based methodology for prostate lesion segmentation in MRI scans using our novel architecture, SAM-UNETR. This architecture is a harmonious blend of the SAM image encoder and a UNET-style transformer decoder. By leveraging the pre-trained weights of complex models, SAM-UNETR provides a powerful yet efficient approach to csPCa detection, combining the richness of transfer learning with the precision of custom segmentation”), further comprising slice selection (Fig. 1 “Prostate cropping process. The network takes a T2w image as input and generates a segmentation mask with three channels: background, CZ, and PZ. The CZ and PZ channels are then combined to form a single mask that represents the whole prostate region … The expanded bounding box is used to crop the prostate region from the T2w image, as well as from the ADC maps and DWI images that correspond to the same slice”, page 118220 Section D. “By conducting segmentation in the 2D space, the risk of losing valuable details on individual image slices is mitigated. This approach safeguards against the potential loss of crucial spatial context and ensures that the segmentation algorithms can effectively capture important features within the lesions”, Fig. 4 n slices, etc.,), and a potential (see 112(b) above) structure model equivalent (page 118222 Section 4) SAM-UNETR “Additionally, a prompt encoder is employed to encode the input prompts into a spatial representation. The mask decoder harmoniously integrates the image features with the prompt representation, effectively generating masks for each prompt in a coherent and precise manner”, Fig. 3, etc.,) if permissible interpretation minimally requires that the set of contoured structures associated with user feedback/interactive segmentation (page 118222, Section 4) SAM-UNETR “The proposed method combines the image encoder from the SAM, a pioneering segmentation model introduced by Kirillov et al. [24], with the decoder architecture inspired by UNETR. … SAM’s training procedure involved an extensive dataset comprising 11 million images and 1.1 billion corresponding masks, called SA-1B. The creation of the dataset involved a unique ‘‘data engine’’ approach that included three stages: assisted-manual, semi-automatic, and fully automatic annotation. In the first phase, the Segment Anything Model (SAM) assists annotators in annotating masks, similar to traditional interactive segmentation”, etc.,) are processed through model architecture/convolutional/up-sampling layers serving to ensure a desired output image resolution and dimensionality is reached (see Fig. 3, SAM-UNETR Architecture, and also relevant decoder disclosure for 3D-UNET for zonal segmentation).
As identified above in the corresponding 112(b) rejection, Applicant may intend for the “structure model” and corresponding set of “contoured structures” to be 3D in nature, however this is not expressly recited (for at least claim 1, claim 9 may differ). Alzate-Grisales discloses, with reference to Fig. 4, 2D models in view of that Transpose block prior to input into each of those four 2D models (for n 2D slices), in a final stage of the methodology (lesion segmentation), in addition to the use of a 3D-UNET architecture to perform preprocessing zonal segmentation discriminating at least central and peripheral zones of the prostate. Alzate-Grisales in further view of page 118220 Section C suggests the obvious nature of model modification to implement an otherwise 2D model/ counterpart as 3D (page 118220 “The 3D-UNET previously mentioned for prostate segmentation is a derivative of the standard UNET architecture. The main difference between the 3D-UNET and the conventional UNET is that the former uses 3D convolutional kernels, as opposed to the 2D counterparts used by the latter. This adaptation allows the 3D-UNET to exploit enhanced spatial information from the input volumes. As described in [17] and Figure 2 shows, the proposed architecture consists of five residual blocks accompanied by five successive downsampling stages”).
It would have been obvious to a person of ordinary skill in the art, before the effective filing date, to modify the system and method of Fuchs as applied to WSI, to be further/additionally applied to a plurality/n slices of a medical image volume as taught/suggested by Alzate-Grisales, and also so as to further implement a “structure model” equivalent as taught/suggested by Alzate-Grisales such that, by means of adjacent slices, and/or surrounding volume information more broadly, related/ segmentation-corroborating spatial information may be considered in the subsequent retraining of Fuchs, similar to that user-provided feedback/annotation information disclosed by Fuchs and interactive/human-in-the loop generation of training datasets for segmentation model training/retraining, also disclosed in Alzate-Grisales and common to SAM.
As to claim 2, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method further comprising testing the personalized segmentation model, using a third set of planning image structures (Fuchs testing/validation datasets (and corresponding structures within imagery), as distinguished from training datasets e.g. [0058] “Thirteen cases may be used for training and the other 42 cases for testing. Note 8 testing cases do not contain the necrosis ratio on their pathology reports, so 34 cases were used for evaluation. Two annotators selected 49 WSIs from 13 training cases and independently annotated them without case-level overlaps”), by determining whether a metric of the personalized segmentation model is within a user variability range (Fuchs [0048] “The experiments show that the CNN model trained by only 7 hours of annotation using DIaL can successfully estimate ratios of necrosis within expected inter-observer variation rate for non-standardized manual surgical pathology task”, [0060], [0028] “The final model, Model2b, achieves the error rate of 20% considered as an expected inter-observer variation rate”, etc.,).
As to claim 3, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 2.
Fuchs suggests the method wherein the metric includes a DICE coefficient or a Fuchs [0028], [0060] “Model2b was selected as the final model because the error rate stopped converging after the second correction. The final model, trained by only 7 hours of annotations done by DIaL, was able to achieve the error rate of 20%, where a 20% inter-observer error rate is considered acceptable for non-standardized tasks in surgical pathology”; see also RDL and RPATH of [0057] and [0060] respectively, also in view of [0070]).
Fuchs fails to explicitly disclose a DICE coefficient or a surface distance error as metric embodiments.
Alzate-Grisales however evidences the obvious nature of a DICE coefficient as an inter-observer variation metric (Abs “This demonstrates the adaptability of large-scale models for different tasks. SAM-UNETR attains a Dice Score of 0.467 and an AUROC of 0.77 for csPCa prediction”, page 118223 Section III “To measure this ability, two common metrics are used: Dice Score and Intersection over Union (IoU) Score. These metrics compare the overlap between the predicted lesion mask and the ground truth lesion mask. The higher the Dice Score and IoU, the better the model’s performance. These metrics are suitable for evaluating segmentation tasks, as they account for both the size and shape of the lesions”).
It would have been obvious to a person of ordinary skill in the art, before the effective filing date, to further modify the system and method of Fuchs in view of Alzate-Grisales so as to implement DICE and/or AUROC metrics for the purpose of evaluating model success, as taught/suggested by Alzate-Grisales, and as any of known/ traditionally implemented success measures (MPEP 2143 Rationale (E)) characterized by a reasonable expectation of success, in further view of their established use in the context of segmentation (page 118223 Section III).
As to claim 4, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method wherein the initial segmentation model is generated using contouring by a set of clinicians of a clinic, the user being a member of the clinic (Fuchs pathologist disclosure of e.g. [0061], [0069] - Examiner does not understand claim 4 to rise to the level of being indefinite (MPEP 2173.05(p)), under any rationale similar to e.g. the findings of In re Katz Interactive Call Processing Patent Litigation, 639 F.3d 1303, 1318, 97 USPQ2d 1737, 1748-49 (Fed. Cir. 2011), since the claim does not recite steps/actions as performed by users interacting with the system and thereby creating confusion as to when direct infringement occurs (claim 1 remains in the context of a ‘receiving’) – but instead limits a model generation via characteristics of an initial dataset used in generating the segmentation model, and “the user being a member of the clinic” is understood to bear little if any patentable weight, since 1) Examiner understands all clinicians to inherently be ‘members’ of at least one ‘clinic’ and 2) the limitation in question requires a non-specific use of a dataset/ contouring created with assistance from one or more (a set) of expert(s)/ physician/clinician/pathologist equivalents – see MPEP 2143.03, in view of those non-exhaustive examples that “may raise question as to limiting effect”; Further, for Examination purposes under BRI, Examiner does not distinguish a pathologist from a clinician (even if it may be argued the former studies disease with potentially less of an emphasis on patient treatment/managing illness)).
Should Applicant assert that pathologist labeling disclosure of Fuchs is limited to a fine-tuning using DIaL and not necessarily a labeling characterizing the initial/pre-trained breast model, Examiner takes Official Notice (MPEP 2144.03) to the manner in which such pre-trained models are generally created based at least in part on expert provided/labeled training samples for supervised learning (Alzate-Grisales’ reference to SAM evidences the same (even if the expert in SAM is not necessarily a clinician equivalent)). Fuchs’ teaching and supporting rationale for utilizing expert annotated samples in a fine-tuning is further applicable to an instance of generating a pre-trained model (even if requiring a much larger dataset and the given increase in cost/time associated with human/expert labeling), and it would have been further obvious to modify Fuchs in view of Alzate-Grisales accordingly (i.e. such that the pre-trained breast model was generated based at least in part on human/expert labeled samples for the breast domain (with no constraints regarding the number of such samples)), recognizing that expert-provided labels are generally high quality (evidencing a reasonable expectation of success), if not a superior ground-truth, particularly for rare/edge cases.
As to claim 5, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method further comprising, sending, to the user, updated slices from the set of contoured structures for iteratively adjusting the updated slices ([0048] “Deep Interactive Learning (DIaL) may be used to minimize pathologists' annotation time by iteratively annotating mislabeled regions to improve a model. DIaL may be used with a pretrained model from a different cancer type to reduce manual training annotation on pancreatic pathology images”, [0051] “by repeating the following three steps: training, segmentation, and correction. These three steps are repeated until annotators are satisfied with segmentation predictions on training images”, [0061] “A pretrained breast model was fine-tuned using DIaL to segment pancreatic carcinomas. During the first iteration, a pathologist annotated false positives on non-tumor subtypes that are not presented on breast training images. The first correction took an hour (example depicted in FIG. 2B). During the second iteration, the pathologist annotated false negatives on pancreatic carcinomas. The second correction took two hours. The pathologist spent total 3 hours to annotate 14 pancreatic pathology images. For numerical evaluation, 23 other images balanced between well-differentiated, moderately differentiated, and poorly differentiated cases were selected and exhaustively annotated by another pathologist”, etc.,).
As to claim 6, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method further comprising, receiving, during a clinical workflow, a user edit to a segmentation generated by the personalized segmentation model, and using the segmentation to iteratively update the personalized segmentation model via the transfer learning technique (see disclosure for e.g. claim 5 above, [0048], [0051], [0061] in further view of permissible interpretation under BRI for a ’clinical workflow’ which may include a pathologist supervised model fine-tuning, and that explicitly disclosed treatment response assessment of Fuchs – e.g. [0023] “This training-segmentation-correction iteration, denoted as Deep Interactive Learning (DIaL), is repeated until segmentation predictions are satisfied by annotators. The final model is used to segment testing WSIs to assess treatment responses”, [0044] “Section A describes systems and methods for deep interactive learning for treatment response assessment”; see also Alzate-Grisales disclosure wherein the segmentation result is disclosed as ‘clinically significant’ (underscoring how it may then be used “in diagnosis, treatment planning, and monitoring of the disease” – Alzate-Grisales Abs)).
As to claim 7, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method wherein the adjusted selected slices include ten or fewer slices (Fuchs discloses in view of [0058] and [0061], various annotation iterations wherein the highest time burden – at 3 hours, was associated with the annotation of 14 pancreatic pathology images (suggesting an average per-image annotation costing 13 minutes); Fuchs falls silent regarding how many images are involved in that first false positive annotation taking 1 hour, and that second correction/annotation taking 2 hours, however it stands to reason giving the average suggested above, that those annotation iterations involved less than 14 images given the 1 and 2 hour times respectively, e.g. closer to 5 and 10 images respectively).
Examiner would further assert that the limitation in question is a design choice constraint, that may be met in view of a motivation to minimize the time any select pathologist/clinician need spend on the annotating/fine-tuning (any number of previous iterations may be performed by an alternative user), and/or as a consequence arising from any specific user deciding that a sufficient number of samples have been annotated/adjusted/corrected (i.e. in view of Fuchs user interface disclosure suggesting that the user may decline to provide additional feedback/annotations, if satisfied with model performance (as a subjective determination) in a recent iteration; see also additionally cited literature directed to ‘few-shot’ instances in the context of DIaL).
It would have been obvious to a person of ordinary skill in the art, before the effective filing date, to further modify the system and method of Fuchs in view of Alzate-Grisales such that the adjusted/corrected/annotated samples/images number 10 or fewer, as taught/suggested by Fuchs in view of that rationale presented above (given the associated 1 hour and 2 hour times spent on the 1st and 2nd iterations – [0061]), and more broadly as an “Obvious to Try” (MPEP 2143 Rationale(s) (E) and/or (G)) number of curated samples, characterized by a reasonable expectation of success (given the prevalence of few and zero-shot strategies in general), the motivation as similarly taught/suggested therein and readily recognized by POSITA that such a concise set of annotations may serve to avoid an expert-time-cost that might otherwise discourage continued fine-tuning iterations and/or use of the model all-together.
As to claim 8, Fuchs in view of a Alzate-Grisales teaches/suggests the method of claim 1.
Fuchs in view of a Alzate-Grisales teaches/suggests the method wherein the initial segmentation model and the personalized segmentation model are specific to a particular disease or organ (see Fuchs various instances of e.g. pancreatic carcinoma disclosure [0048], [0061], and that initial pre-trained breast model [0061], also Alzate-Grisales clinically significant prostate cancer (csPCa) – for clarity of record purposes, Examiner does not understand claim interpretation under BRI to require initial and personalized models to be specific to the same organ and/or disease (e.g. implying no constraints on whether the TL is cross-disease, cross-organ, or concerning some other domain difference)).
As to claim 10, this claim is the CRM (see 101 rejection above – regarding Step 1 analysis and transitory signal and/or program per se ‘medium’ embodiments) claim corresponding to the method of claim 1 and is rejected accordingly.
As to claims 11-17, these claims are the CRM claims corresponding to method claims 2-8 respectively, and are rejected accordingly.
As to claim 19, this claim is the system claim corresponding to and sufficiently incorporating the scope of the computer implemented method of claim 1, and is rejected accordingly. Fuchs presents multiple instances of corresponding process and memory disclosure (see e.g. Fig. 8, [0132-0135], etc.,).
As to claim 20, this claim is the system claim corresponding to the method of claim 8 and is rejected accordingly.
Additional References
Prior art made of record and not relied upon that is considered pertinent to applicant's disclosure:
Additionally cited references (see attached PTO-892) otherwise not relied upon above have been made of record in view of the manner in which they evidence the general state of the art.
Allowable Subject Matter
Claims 9 and 18 would be allowable if rewritten to overcome the rejection(s) under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), 2nd paragraph, set forth in this Office action and to include all of the limitations of the base claim and any intervening claims. References of record fail to serve in any obvious combination teaching each and every limitation as required therein, particularly in view of the previously proposed combination of references as required for the proper rejection of independent claim(s) and modification(s) thereto in the aggregate as would be necessary.
Inquiry
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IAN L LEMIEUX whose telephone number is (571)270-5796. The examiner can normally be reached Mon - Fri 9:00 - 6:00 EST.
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, Chan Park can be reached on 571-272-7409. 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.
/IAN L LEMIEUX/Primary Examiner, Art Unit 2669
Prosecution Insights