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 .
Response to Amendment
The amendment filed on 6/10/2026 has been entered. Claims 1-20 remain pending in the application. Applicant’s amendments to the Specification and Claims have overcome each and every objection previously set forth in the Non-Final Office Action mailed 4/01/2026.
Response to Arguments/Remarks re. 35 U.S.C. § 102/103 Rejections
Applicant’s arguments with respect to image augmentation have been considered but are moot because the new ground of rejection does not rely solely on Zheng for any teaching or matter specifically challenged in the argument.
Applicant argues that Zheng does not teach claim 1 as “each of the second plurality of data is not paired with another image”. Applicant also states that “it should be noted that augmentations of a same image is characteristically different from image pair with independently acquired images.” The examiner understands such an augmentation as producing a “pair” (particularly when these images are later compared against each other as in Zheng, paragraph 72). It appears the applicant may have a particular meaning for the word pair, but that meaning is not explicitly included within the claim language so as to prevent such a reading of the claim.
The applicant further argues that Zheng does not have a "precision loss term". However, that phrase is not present in the language of claim 1, and therefore the argument is moot. The plain language as written is met by paragraphs 72-74 of Zheng as previously cited, which incorporates term Lc (computed under paragraph 71) that takes a difference between y1 and y2 (predicted BMDs corresponding to the image pair).
Examiner’s Note
The examiner considered under MPEP § 2173.05(i) whether the amended language of claim 1 constituted an impermissible negative limitation: “and wherein the first unlabeled image and the second unlabeled training image are not two augmentations of a same image”. The examiner concluded that the language is sufficiently supported (second paragraph of page 3 of the claimed invention’s specification):
“In one embodiment, to ensure a similarity in the main, in each of the multiple secondary learning data the first unlabeled training image and the second unlabeled training image are images of the same subject taken within a predetermined time interval, wherein within the predetermined time interval the main feature is known to be constant or only varies within the measurement limit of the measuring method. Specifically, the predetermined time interval may be 3 months or 6 months”
As the images are taken within a predetermined time interval, it naturally follows that in a certain embodiment they are not augmentations of the same image.
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, 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.
Claims 1-3, 5-6, 8-10, 12, 14, and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng et. al (US 20220309651 A1) (Hereinafter, “Zheng”) in view of Dwibedi et. al With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations (Hereinafter, “Dwibedi”)
With respect to claim 1, Zheng teaches:
A method of training a prediction model to generate one or more main predicted values of a main feature of an input image (Fig. 1), comprising training the prediction model with a primary dataset ([0004] fine-tuning the pre-trained model on a first plurality of data representing the labeled one or more ROIs) and a secondary dataset ([0004] and a second plurality of data representing unlabeled region) by adjusting multiple parameters in the prediction model ([0079]) to lower a total loss of the prediction model ([0073]); wherein: the primary dataset comprises multiple primary learning data, each of which comprises a labeled training image labeled with one or more main ground truth values of the main feature (Fig. 4; [0019] “For example, paired hip X-ray image and DEXA measured BMD are collected as labeled data for supervised regression learning”; [0035]-[0036])
the secondary dataset comprises multiple secondary learning data, each of which comprises an unlabeled training image pair containing a first unlabeled training image and a second unlabeled training image having similarity in the main feature ([0071]-[0074])
the total loss comprises a primary loss and a secondary loss ([0073])
the primary loss is calculated based on the difference between the one or more main ground truth values and one or more primary predicted values of the labeled training image (Fig. 4; [0034] “Step 401: Determining a mean square error (MSE) loss between an estimated BMD and a ground-truth (GT) BMD”), the one or more primary predicted value are one or more values of the main feature generated by the prediction model ([0034])
the secondary loss is calculated based on the difference between one or more first predicted values of the first unlabeled training image and one or more second predicted values of the second unlabeled training image ([0072]-[0074]), the one or more first predicted values and the one or more second predicted values are one or more values of the main feature generated by the prediction model ([0072])
Zheng does not explicitly teach:
wherein the first unlabeled training image and the second unlabeled training image are not two augmentations of a same image
However, Dwibedi, in the same field of machine-learning implemented image analysis, teaches:
A method of training a prediction model to generate one or more main predicted values of a main feature of an input image ([4.1]) comprising training the prediction model with a [3.2]; [4.1]; [4.3])
wherein the [4.1] “We train our NNCLR representation on the ImageNet2012 dataset…without using any annotation or class labels”) containing a first unlabeled training image and a second unlabeled training image having similarity in the main feature, and wherein the first unlabeled training image and the second unlabeled training image are not augmentations of a same image (Fig. 2 “Overview of NNCLR Training”; [1] “While most methods treat different views of the same image as positives for a contrastive loss, we are interested in using positives from other instances in the dataset”; [3.2] “Instead, we propose using z1 nearest-neighbor in the support set Q to form the positive pair”; [4.1])
the [3.2])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng to include the limitations of unaugmented pair learning as taught by Dwibedi. Doing so would have the advantage of training the model to learn conceptually and avoid overfitting. The systems readily integrate, as the augmented training images of Zheng can be readily substituted with similar, but different images under the principle of Dwibedi without any compromise to underlying functionality, as the modification from Dwibedi represents a different training input rather than an alteration to core structure.
With respect to claim 2, Zheng/Dwibedi teaches:
(Noting that the remainder of the citations under this combination refer to disclosure in Zheng)
The method of claim 1, further comprising training the prediction model to generate one or more auxiliary predicted values of one or more auxiliary features of the input image by adjusting the multiple parameters in the prediction model to lower the total loss of the prediction model ([0040]-[0041] equations 2-4; [0071]-[0073] equation 6, noting that feature term and BMD term are distinct), wherein:
the one or more auxiliary features are correlated with the main feature ([0006] “The one or more loss functions includes a specific adaptive triplet loss (ATL) configured to encourage distances between one or more feature embedding vectors correlated to differences among the BMDs”)
the labeled training image of each of the multiple primary learning data is further labeled with one or more auxiliary ground truth values of the one or more auxiliary features (Fig. 3; [0026]-[0032])
PNG
media_image1.png
946
1195
media_image1.png
Greyscale
the total loss further comprises a tertiary loss ([0030]; [0035]-[0044]; Fig. 3; Fig. 4, discussing computation of adaptive triplet loss, which necessarily utilizes the loss of [30]’s supervised pre-training for the purpose of generating accurate feature embeddings that are then correlated with BMD, adaptive triplet loss then being used in [0073] to compute total loss)
the tertiary loss is calculated based on the difference between the one or more ground truth values and one or more tertiary predicted values of the labeled training image, the one or more tertiary predicted values are one or more values of the one or more auxiliary features generated by the prediction model (Fig. 3; [0030])
With respect to claim 3, Zheng/Dwibedi teaches:
The method of claim 1, wherein the labeled training image is modified by image augmentation before generating the one or more primary predicted values by the prediction model (Fig. 3; [0026] “Random affine transformations, color jittering, and horizontal flipping may also be applied to resized ROI during training”)
With respect to claim 5, Zheng/Dwibedi teaches:
The method of claim 1, wherein the primary and the secondary loss are calculated by squared loss function ([0036] equation 1; [0071] equation 6)
With respect to claim 6, Zheng/Dwibedi teaches:
The method of claim 1, wherein in each of the multiple secondary learning data the first unlabeled training image and the second unlabeled training image are images of the same subject taken within a predetermined time interval to have similarity in the main feature ([0026] “The X-ray images may be taken within six months of the BMD measurement.”)
With respect to claim 8, Zheng/Dwibedi teaches:
The method of claim 1, wherein each of the labeled training image, the first unlabeled training image and the second unlabeled training image is an ROI (region of interest) extracted image extracted from an original training image via ROI extraction ([0026])
With respect to claim 9, Zheng/Dwibedi teaches:
The method of claim 1, wherein each of the labeled training image, the first unlabeled training image and the second unlabeled training image is a training image set comprising:
an original training image: and
an ROI (region of interest) extracted image extracted from the original training image via ROI extraction ([0026] “In some embodiments, to extract ROI images around the femoral neck, an automated ROI localization model may be trained with the deep adaptive graph (DAG) network using about 100 images with manually annotated anatomical landmarks; [0032] “As shown in Fig. 3, in the self-training stage illustrated by step 304, the model may be fine-tuned on two groups of data. The two group of data includes a first plurality of data which represent the labeled ROI image and a second plurality of data which represent unlabeled region”; Fig. 3)
With respect to claim 10, Zheng/Dwibedi teaches:
The method of claim 1, wherein the main feature is bone density of a subject, and the one or more main predicted value are one or more bone mineral density (BMD) values ([0036])
With respect to claim 12, Zheng/Dwibedi teaches:
The method of claim 10, wherein the labeled training image in each of the primary learning data, and the first unlabeled training image and the second unlabeled training image in each of the secondary learning data are X-ray images ([0026])
With respect to claim 14, Zheng/Dwibedi teaches:
The method of claim 10, further comprising training the prediction model to generate one or more auxiliary predicted values of one or more auxiliary features of the input image by adjusting the multiple parameters in the prediction model to lower the total loss of the prediction model ([0040]-[0041] equations 2-4; [0071]-[0073] equation 6, noting that feature term and BMD term are distinct), wherein:
the one or more auxiliary features are correlated with the bone density of a subject ([0006] “The one or more loss functions includes a specific adaptive triplet loss (ATL) configured to encourage distances between one or more feature embedding vectors correlated to differences among the BMDs”)
the labeled training image of each of the multiple primary learning data is further labeled with one or more auxiliary ground truth values of the one or more auxiliary features (Fig. 3; [0026]-[0032])
the total loss further comprises a tertiary loss ([0030]; [0035]-[0044]; Fig. 3; Fig. 4, discussing computation of adaptive triplet loss, which necessarily utilizes the loss of [30]’s supervised pre-training for the purpose of generating accurate feature embeddings that are then correlated with BMD, adaptive triplet loss then being used in [0073] to compute total loss)
the tertiary loss is calculated based on the difference between the one or more auxiliary ground truth values and one or more tertiary predicted values of the labeled training image, the one or more tertiary predicted values are one or more values of the one or more auxiliary features generated by the prediction model (Fig. 3; [0030])
With respect to claim 17, Zheng/Dwibedi teaches:
The method of claim 12, wherein the labeled training image is a training image set comprising:
an original training image
an ROI extracted image which is an identified ROI region of a hip joint extracted from the original training image ([0018] “As used herein, the term “hip X-ray” refer to X-ray imaging results and/or X-ray examinations, that can help to detect bone cysts, tumors, infection of the hip joint, or other diseases in the bones of the hips, etc.”; [0026]; [0032]; Fig. 3)
With respect to claim 18, Zheng/Dwibedi teaches:
The method of claim 17, wherein the original training image and the ROI extracted image are modified by image augmentation before generating the one or more primary predicted values by the prediction model ([0026]; [0071]-[0073] With respect to the original training image, it is also augmented for the purpose of minimizing total loss in the system as a whole. This is done “before” generating the one or more primary predicted values, because it takes place in a fine-tuning stage as detailed in [0058])
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Pinkovich et. al (US 20210034921 A1) (Hereinafter, “Pinkovich”).
With respect to claim 4, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the one or more main ground truth values are modified by ground truth augmentation before calculating the primary loss
However, Pinkovich in the same field of endeavor of medical imaging and machine learning, teaches:
wherein the one or more main ground truth values are modified by ground truth augmentation before calculating the primary loss ([Abstract]; [0090]; “On the other hand, geometric transformations such as translation, rotation, zoom or even non-rigid transformations may need to be applied to the ground truth coordinates in the same way they are applied to the image in order to maintain validity of the ground truth coordinates”; [0104])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of ground truth augmentation as taught by Pinkovich. As Pinkovich teaches, ground truth augmentation increases the robustness of a machine learning model by reducing the probability of overfitting [0004]. Zheng/Dwibedi and the claimed invention both benefit from a stronger initial training dataset, and such a system is readily compatible with the base components of Zheng/Dwibedi.
Claims 7 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Sjöstrand et. al (US 20250104225 A1) (Hereinafter, “Sjöstrand”).
With respect to claim 7, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the predetermined time interval is 3 months
As Zheng/Dwibedi teaches images collected within a 6 month time span, but not explicitly 3 months (Zheng, [0026])
However, Sjöstrand, in the same field of endeavor of medical image analysis and machine learning, teaches:
wherein the predetermined time interval is 3 months ([0034] “the one or more medical images are obtained within six (6) months or less (e.g., three months or less)”; [0099])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of explicit 3-month collection as taught by Sjöstrand. Sjöstrand’s methods apply to machine learning and more specifically bone-scan and x-ray analysis ([0099]), and as such would be consulted by one wishing to develop machine learning methods in the biomedical space. One of ordinary skill in the art would be motivated to introduce a 3-month collection model for the advantage of more rapid detection of change by a machine-learning model. Zheng/Dwibedi already encourages the collection of images within this time span (Zheng, [0026]) and as such the teachings are readily integrated.
With respect to claim 13, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the first unlabeled training image and the second unlabeled training image are two X-ray images of the same subject taken sequentially within 3 months
As Zheng/Dwibedi teaches X-ray images collected within a 6 month time span, but not explicitly 3 months (Zheng, [0026])
However, Sjöstrand teaches:
wherein the first unlabeled training image and the second unlabeled training image are two X-ray images of the same subject taken sequentially within 3 months ([0034]; [0099] “Examples of anatomical images include, without limitation x-ray images…”)
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of explicit 3-month collection as taught by Sjöstrand. Sjöstrand’s methods apply to machine learning and more specifically bone-scan and x-ray analysis ([0099]), and as such would be consulted by one wishing to develop machine learning methods in the biomedical space. One of ordinary skill in the art would be motivated to introduce a 3-month collection model for the advantage of more rapid detection of change by a machine-learning model. Zheng/Dwibedi already encourages the collection of images within this time span ([0026]) and as such the teachings are readily integrated.
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Vlachos et. al Is Regional Bone Mineral Density the Differentiating Factor Between Femoral Neck and Femoral Trochanteric Fractures? (Hereinafter, “Vlachos”)
With respect to claim 11, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the one or more main predicted values comprise bone mineral density (BMD) values of total hip, femoral neck, greater trochanter, and femoral shaft
However, Vlachos, in the same field of endeavor of bone mineral density measurement, teaches the limitations of:
wherein the one or more main predicted values comprise bone mineral density (BMD) values of total hip, femoral neck, greater trochanter, and femoral shaft (Table 2, disclosing measurements of BMD for each area)
It would have been obvious to one of ordinary skill as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of specific BMD measurement as taught by Vlachos. Doing so would allow Zheng/Dwibedi to greater distinguish between specific regions and provide a more descriptive output dataset. The methods of Vlachos and Zheng/Dwibedi are readily compatible, as Zheng/Dwibedi already considers a broader class of BMD values which are capable of being narrowed with predictable success.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Arnaud et. al (US 20110040168 A1) (Hereinafter, “Arnaud”).
With respect to claim 15, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the one or more auxiliary features comprise cortical thickness of the subject, and the one or more auxiliary predicted values comprise a cortical thickness index (CTI) value of the subject
As Zheng/Dwibedi teaches features correlated to bone mineral density generally (Zheng, [0006])
However, Arnaud, in the same field of endeavor of bone disease prediction teaches:
wherein the one or more auxiliary features comprise cortical thickness of the subject, and the one or more auxiliary predicted values comprise a cortical thickness index (CTI) value of the subject ([0131])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of CTI as taught by Arnaud. Doing so would provide a more specific correlating feature which could be used to execute the method of Zheng/Dwibedi taught more broadly. One of ordinary skill in the art would readily consult Arnaud as it also seeks to predict bone disease using specific correlating features.
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Fouts et. al (US 20200253667 A1) (Hereinafter, “Fouts”).
With respect to claim 16, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the one or more auxiliary features comprise femoral neck width of the subject, and the one or more auxiliary predicted values comprise a femoral neck width (FNW) value of the subject
As Zheng/Dwibedi teaches features correlated to bone mineral density generally, including the femoral neck, but not width specifically (Zheng, [0006]; Zheng, [0025])
However, Fouts, in the same field of endeavor of bone disease prediction, teaches:
wherein the one or more auxiliary features comprise femoral neck width of the subject, and the one or more auxiliary predicted values comprise a femoral neck width (FNW) value of the subject ([0077])
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the limitations of FNW as taught by Fouts. Doing so would provide a more specific correlating feature which could be used to execute the method of Zheng/Dwibedi taught more broadly. One of ordinary skill in the art would readily consult Fouts as it also seeks to predict bone disease using specific correlating features.
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Asiedu et. al (US 20210374953 A1) (Hereinafter, “Asiedu”) and Laserson et. al (US 20220318565 A1) (Hereinafter, “Laserson”).
With respect to claim 19, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein said image augmentation is performed by cropping 0-25% of the original training image without cropping the identified ROI region, and wherein said image augmentation is performed by shifting the identified ROI region by 0-7% in a specific direction.
As Zheng teaches cropping and image augmentation generally, but not in any precise amounts (Zheng, [0026])
However, Asiedu, in the same field of endeavor of image analysis and disease prediction teaches:
wherein said image augmentation is performed by cropping 0-25% of the original training image without cropping the identified ROI region ([0056] “Due to image positioning diversity which for each cervigram, the cervix region of interest (ROI) was cropped using a minimum bounding box around the cervix region identified by an expert colposcopist. With standardized images in which the cervix took up about 90% of the image, no cropping was necessary” noting that a crop of 0% reads on the claim under the broadest reasonable interpretation standard)
And Laserson, in the same field of endeavor of image analysis and disease prediction teaches:
wherein said image augmentation is performed by shifting the identified ROI region by 0-7% in a specific direction ([0180]-[0182] “The following preprocessing and augmentation steps were applied to each patch: Add random noise to the Top and Bottom coordinates, up to ±5% of box height. [0182] Add random noise to the Left and Right coordinates, up to ±5% of box width.”)
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include the specific means of image augmentation as taught by Asiedu and Laserson. Doing so would allow a person of ordinary skill in the art to specifically execute the means of Zheng/Dwibedi using reference amounts already disclosed in the art. Such teachings are readily compatible as Asiedu and Laserson also introduce these means for the purposes of detecting pathologies in image analysis.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Zheng/Dwibedi in view of Leuliet Deep learning for tomographic reconstruction: Study and application to computed tomography and positron emission imaging (Hereinafter, “Leuliet”)
With respect to claim 20, Zheng/Dwibedi does not explicitly teach the limitations of:
wherein the one or more main ground truth values are modified by introducing small variables randomly selected between -0.01 g/cm2 and 0.01 g/cm2
However, Leuliet, in the same field of endeavor of bone mineral density prediction teaches:
wherein the one or more main ground truth values are modified by introducing small variables randomly selected between -0.01 g/cm2 and 0.01 g/cm2 ([2.2]; [4.2.5] discussing injection of gaussian noise with standard deviation of 1% mean HU value [4.2.4.3] discussing estimation of BMD from HU values; [4.1.2] discussing BMD in g/cm2)
Zheng/Dwibedi takes a sample population of patients, who have average BMD values well below 1.0 g/cm2 (Zheng, [0026]; Zheng, Fig. 7). As such, if one were to incorporate the teachings of Leuliet (to introduce a variance equal to 1% of the mean value), the result would be a variance within the range of the claim language.
It would have been obvious to one of ordinary skill in the art as of the effective filing date of the claimed invention, to modify Zheng/Dwibedi to include augmentation of ground truth data within a 1% range as taught by Leuliet. Doing so would ensure the data greater mimics real-world data ([2.2]), increasing the effectiveness of the training. Zheng/Dwibedi already collects a dataset of ground truth BMD values, for which noise in this range could be readily injected prior to training, in the recommended amount of 1% mean (thus necessarily resulting in the values of the claimed language).
Conclusion
Applicant’s amendment necessitated the new grounds 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 NOAH WILLIAM BOYAR whose telephone number is 571-272- 8392. The examiner can normally be reached 10:00 AM – 6:00 PM EST, Monday – Friday. 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 at 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.
/NOAH W BOYAR/
Examiner, Art Unit 2669
/CHAN S PARK/Supervisory Patent Examiner, Art Unit 2669
Prosecution Insights