|
View Edit Delete |
PER-12 |
Deep neural networks (DNNs) can be one or more ANNs with multiple hidden layers between the input and output layers. DNNs can model complex non-linear relationships. DNNs can generate models where the object is expressed as a layered composition. DNNs can be feedforward networks in which data flows from the input layer to the output layer without looping back, DNNs can be recurrent neural networks or convolutional deep neural networks. |
195 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data can more abundant than labeled data. For example, in a clinical environment, a deep learning system with a multi-layered ANN can initially be trained using a classification of outcomes in a supervised fashion. As the data grow, the system can optionally learn in an unsupervised manner, for example by utilizing pattern recognition across large clinical datasets, which can include pre-operative, intra-operative and post-operative data. (See US Patent 11,278,413 Para. 76-79). |
196 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Classification |
197 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Classification can be a process of creating categories, in which data or objects can be recognized, differentiated or understood. A classification system can be an approach of accomplishing classification. Classification can be performed using mathematical classification, statistical classification, classification theorems, e.g. in mathematics, and attribute value systems. Classifications can be alphanumeric. Classifications can be single or multi-dimensional. Classifications can be single or multi-layered. Classifications can be color coded. An ANN can use a single classification system, e.g. in supervised learning. An ANN can use multiple classification systems. When multiple classification systems are used, they can reside in different layers of a DNN or deep learning system. |
198 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Classification can be the problem of identifying to which of a set of categories or sub-populations a new observation belongs; this can be determined, for example, using a training data set with observations whose category membership is known. For example, a diagnosis can be assigned to a patient as a category which can be described by measured data or characteristics such a heart rate, blood pressure, presence of absence of symptoms or combinations of symptoms. Classification can be a pattern recognition. |
199 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Individual observations or data can be divided into a set of quantifiable properties. These properties may be categorical, e.g. “a”, “b”, “c”, “d” etc. or ordinal, e.g. “excellent”, “very good”, “good”, “acceptable”, “average”, or “poor”. They can be integer or real valued. Observations or data can also be classified using similarity or distance functions, e.g. based on earlier observations or data. An algorithm that implements classification can be a classifier, A classifier can sometimes also be a mathematical function, e.g. implemented by a classification algorithm, that can map input data to a category. |
200 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Data can be classified, for example, into “excellent”, “good”, “acceptable” or “poor” outcome, e.g. one or more clinical outcomes or clinical outcome variables. Numeric values or ranges of numeric values can be assigned to different classes, for example numeric values or ranges of numeric values from a patient reported outcome measurement, from a clinical reporting system, and/or from one or more electronic measurements. |
201 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Discriminative vs. Generative Models and Networks |
202 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In machine learning, discriminative models can be distinguished from generative models. |
203 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
1. Discriminative models are trained to learn the boundaries between classes. They model the conditional probability of a target variable Y (class), given an observation x: P(Y|X=x) (“probability of Y given X=x”). Discriminative models describe the probability for classifying a given example x into a classy E Y. Discriminative models include, for example, logistic regression, conditional random fields, support vector machines, neural networks, random forests, or perceptrons. |
204 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Generative models model the distribution of individual classes. They can generate data and provide a statistical model of the joint probability distribution on X×Y, P(X,Y)=P(X|Y)*P(Y), for an observable variable X and a target variable Y. Generative models include, for example, naïve Bayes models and Bayes networks, hidden Markov models, Boltzmann machines, variational autoencoders or generative adversarial networks (GAN). |
205 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In some embodiments, the computer system can use a trained artificial neural network (ANN) to determine the treatment plan. The ANN can implement a discriminative model. A discriminative model can be trained to classify the input data, i.e. the preoperative and/or intraoperative data and/or postoperative data, into different classes, wherein each class can represent a different treatment plan. |
206 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In some embodiments, the ANN can implement a generative model. Instead of assigning preoperative and/or intraoperative input data and/or postoperative data to an existing class, a generative model is trained to generate the treatment plan steps based on the input data. |
207 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In some embodiments, a generative and a discriminative network model can be combined into a generative adversarial network (GAN) to generate a treatment plan. Using a training data set of existing recorded treatment plans for a number of preoperative and/or intraoperative input data and/or postoperative data sets, in this situation, the generative network can be trained to generate a preferred treatment plan from the preoperative and/or intraoperative input data. The discriminative network can be trained to evaluate the generated treatment plan and to distinguish the generated treatment plan from the actual treatment plan of the training case. Thus, the discriminative network can force the generative network to improve its results. (See US Patent 11,278,413 Para. 80-91). |
208 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Figure 11 illustrates a flowchart diagram depicting a method 1100 for generating one or more patient specific instruments configured to correct a bone condition, according to one embodiment. Prior to steps of the method 1100, a bone model (also referred to as CAD model above) is generated. The bone model may be generated using medical imaging of a patient’s foot and may also be referred to as an anatomic model. The medical imaging image(s) may be used by computing devices to generate patient imaging data. The patient imaging data may be used to measure and account for orientation of one or more structures of a patient’s anatomy. In certain embodiments, the patient imaging data may serve or be a part of anatomic data for a patient. |
209 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In one embodiment, the method 1100 begins after a bone model of a patient’s body or body part(s) is generated. In a first step 1102, the method 1100 may review the bone model and data associated with the bone model to determine anatomic data of a patient’s foot. |
210 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
After step 1102, the method 1100 determine 1104 a deformity in the patient’s anatomy using the anatomic data. In certain embodiments, the detection and/or identification of a deformity may employ advanced computer analysis system, expert systems, machine learning, and/or automated/artificial intelligence. As used herein, "artificial intelligence" refers to intelligence demonstrated by machines, unlike the natural intelligence displayed by humans and animals, which involves consciousness and emotionality. The distinction between artificial intelligence and natural intelligence categories is often revealed by the acronym chosen. 'Strong' AI is usually labelled as artificial general intelligence (AGI) while attempts to emulate 'natural' intelligence have been called artificial biological intelligence (ABI). Leading AI textbooks define the field as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of achieving its goals. The term "artificial intelligence" can also be used to describe machines that mimic "cognitive" functions that humans associate with the human mind, such as "learning" and "problem solving". (Search "artificial intelligence" on Wikipedia.com June 25, 2021. CC-BY-SA 3.0 Modified. Accessed June 25, 2021.) Various kinds of deformities may be identified, such as a bunion. The deformities determined may include congenital as well as those caused by injury or trauma. |
211 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
Next, the method 1100 proceeds and a preliminary cutting guide model is provided 1106 from a repository of template cutting guide models. A preliminary cutting guide model is a model of a preliminary cutting guide. |
212 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
As used herein, "preliminary cutting guide" refers to a guide configured, designed, and/or engineered to serve as a template, prototype, archetype, or starting point for creating, generating, or fabricating a patient-specific cutting guide. In one aspect, the preliminary cutting guide may be used, as-is, without any further changes, modifications, or adjustments and thus become a patient-specific cutting guide. In another aspect, the preliminary cutting guide may be modified, adjusted, or configured to more specifically address the goals, objectives, or needs of a patient or a surgeon and by way of the modifications become a patient-specific cutting guide. The patient-specific cutting guide can be used by a user, such as a surgeon, to guide making one or more resections of a structure, such as a bone for a procedure. Accordingly, a preliminary cutting guide model can be used to generate a patient-specific cutting guide model. The patient-specific cutting guide model may be used in a surgical procedure to address, correct, or mitigate effects of the identified deformity and may be used to generate a patient-specific cutting guide that can be used in a surgical procedure for the patient. |
213 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |
|
View Edit Delete |
PER-12 |
In certain embodiments, the preliminary cutting guide model may be generated based on anatomic data and/or a bone model or a combination of these, and no model or predesigned structure, template, or prototype. Alternatively, or in addition, the preliminary cutting guide model may be, or may originate from, a template cutting guide model selected from a set of template cutting guide model. Each model in the set of template cutting guide models may include one or more cutting resection features positioned and/or sized and/or configured to fit for an average patient’s foot. The template cutting guide model may subsequently be modified or revised by an automated process or manual process to generate the preliminary cutting guide model used in this disclosure. |
214 |
Added by DJM Jan 2024 |
1/6/24, 10:03 PM |