LU506274B1 - A SYSTEM FOR MONITORING DISEASE PROGRESSION IN LEUKEMIA BASED ON THE REGULATION OF HO-1 EXPRESSION - Google Patents

Abstract

The present invention proposes a system for monitoring disease progression in leukemia based on the regulation of HO-1 expression, and at the core of which is a well-designed multimodal deep learning neural network. The system integrates genomic data, clinical data and HO-1 expression levels and effectively analyzes these complex medical data through a multi-head attention mechanism. This approach greatly improves the accuracy and personalization of leukemia treatment and monitoring and provides strong data support for clinical decision-making.

Description

A system for monitoring disease progression in leukemia based on ddr/506274

Regulation of HO-1 expression

Technical area

The present invention relates to the field of medical and health data processing and disease prediction and is particularly directed to the

Monitoring and management of acute myeloid leukemia (AML).

More specifically, the present invention relates to a system for monitoring the

disease progression of leukemia, which combines multimodal deep learning techniques with

Expression data of heme oxygenase-1 (HO-1) were used.

Technology in the background

Acute myeloid leukemia (AML) is a common hematological malignancy characterized by the presence of large numbers of immature white blood cells in the bone marrow. The treatment and management of AML is a complex process that requires comprehensive consideration of data from a variety of aspects, including the patient's genetic information, clinical presentation, and

Laboratory test results. In the treatment of AML, a timely and accurate evaluation of the

Disease progression, response to therapy and patient prognosis are crucial for improving treatment outcomes.

Currently, leukemia surveillance relies mainly on the analysis of

Patients' medical records by clinicians and regular laboratory tests. However, this traditional approach has its limitations: firstly, manual analysis of the data is time-consuming and prone to errors, and secondly, it does not allow real-time monitoring of

Changes in the disease and thirdly, it is difficult to predict the long-term development of

to accurately predict disease.

In recent years, the development of mechanization and technological techniques, especially in the field of data analysis in healthcare, has opened up new opportunities to solve these problems. With the help of these advanced

Data analysis techniques can effectively extract valuable information from complex health data, improving the accuracy and efficiency of disease monitoring. In particular, multimodal deep learning techniques that can process and integrate data from different sources (e.g. genomic data, clinical data, etc.) provide a more comprehensive perspective for the prediction and treatment of leukemia.

A comprehensive system that combines these advanced technologies for the

However, existing technologies lack a prognostically relevant monitoring system that can detect the expression level of HO-1, a biomarker that is important for the

Treatment and prognostic assessment of AML is important.

Therefore, there is a need to develop a new system capable of integrating multimodal data and applying deep learning techniques, especially in

Connection with the HO-1 expression level to improve the accuracy and efficiency of

To improve monitoring of disease progression in AML.

Content of the invention

The present invention provides a system for monitoring disease progression in

Leukemia based on the regulation of HO-1 expression and aims to provide a comprehensive, efficient and precise solution for the diagnosis, monitoring and treatment of acute myeloid leukemia (AML) by applying advanced multimodal deep¥206274

Learning techniques are used. The system combines multimodal data such as genomic data, clinical data, laboratory test results and HO-1 expression values to

Disease progression, response to treatment and patient prognosis by its

Deep learning model to effectively predict.

The system architecture of the present invention comprises the following main modules: 1. Data interface module: Direct interfaces to hospital information systems,

Gene sequencing platforms and laboratory equipment to analyze patient records, genomic data,

laboratory test results, etc. in real time; supports a variety of medical

data formats to ensure data integrity and accuracy. 2. Data processing and integration module: pre-processing of collected data, including data cleaning (removing errors and missing values), conversion (standardization of data formats) and integration (unification of information from different data sources); special attention is paid to the fusion of multimodal data to ensure the effective integration of data into the model. 3. Centralized data warehouse: designed as a powerful data storage solution to ensure secure and fast data access; supports big data technologies to

to facilitate processing and analysis of large amounts of data. 4. Advanced feature engineering module: performs in-depth data analysis to extract key features relevant to the diagnosis and treatment of leukemia, with a particular focus on automatic feature learning capabilities. 5. Multimodal deep learning module: builds neural network models with multiple attention mechanisms and complex feature extraction layers suitable for

leukemia. 6. Decision support system: providing data-driven

Decision support for doctors and patients using the results of the deep learning

model, including disease prediction, treatment recommendation and prognosis assessment. 7. Interactive user interface: Providing an intuitive user interface for

Doctors and patients to query information, view prediction results and provide feedback. 8. Continuous learning and optimization module: Continuously optimizes the

Model performance and accuracy based on user feedback and newly collected

Data. 9. System Management and Monitoring Module: Responsible for daily operations and maintenance, including performance monitoring, security management and

Data backup.

The workflow of the present invention is as follows: 1. Data acquisition and reception

The system has interfaces to several data sources, such as electronic

Medical records in hospitals, gene sequencing laboratories, etc. to provide real-time detailed patient medical data, genetic test results, blood analysis data and specific

to receive biomarker data, including HO-1 expression levels.

This data can be divided into four broad categories:

A. Genomic data: sequencing data of the patient's entire genome, including gene variants, gene expression patterns, etc.; targeted sequencing results for specific genetic markers, e.g. for specific genes associated with AML. LUS06274

B. Clinical data: patient medical records, including diagnostic information,

Medical history, results of physical examination, treatment records, including

Drug regimen, dosage and response; laboratory test results, such as hematological

Tests, biochemical indicators, etc.

C. Time series data: Patient follow-up data, including clinical

Indicators, response to treatment and recording of symptoms at regular examinations; physiological measurements over time, such as blood pressure,

blood sugar etc.

D. HO-1 expression data: HO-1 protein and gene expression levels, which are usually determined by

Blood tests or tissue biopsies; correlation data between HO-1 expression and disease progression, treatment response. 2. Data preprocessing and quality control

A. Data cleansing: identification and correction of erroneous data, treatment of

Outliers and filling missing values. To ensure data integrity, advanced algorithms are used to estimate and interpolate missing data.

B. Data standardization and normalization: Conversion of data from different

sources into a uniform format and scope to reduce the impact of

Reduce variability of data sources.

C. Data validation: Quality control of data to ensure the accuracy and reliability of the input data. 3. Feature engineering and data integration

A. Methods for feature extraction: For genomic data, the

Principal component analysis (PCA) and gene expression network analysis are used to identify important gene variants and expression patterns; for clinical data,

Time series analysis and trend analysis algorithms are used to identify changes in clinical

metrics; and for drug response data, drug response network modeling and pharmacodynamic modeling are used to predict patient response to specific

to evaluate and predict medications.

B. Data integration techniques: Structured data (e.g., laboratory test results) are integrated using standardized processing (e.g., unit conversion, range adjustment); unstructured data (e.g., physician notes, imaging reports) are integrated using

Natural language processing (NLP) techniques such as named entity recognition and

Sentiment analysis converted into structured formats; data from various sources and

Formats are effectively integrated through collaborative filtering and hybrid models.

C. Feature selection and optimization: Machine learning algorithms such as Random

Forest and Gradient Boosting Decision Tree are applied to select the most informative features based on the model's feature importance evaluation;

Correlation analyses are performed on the feature set to reduce redundant features, and the feature set is further optimized by L2 regularization methods.

With the approach described above, the system is able to identify the most valuable

Extract information from the rich and complex multimodal data and integrate it into a comprehensive picture of the state that provides the deep learning model with high-quality

provides inputs.

4. Training and optimization of the deep learning model LUS06274

The Multi-Head Attention mechanism allows the model to focus on the most important

information in the data, improving its ability to capture complex patterns. The model is tuned using a validation set, including adjusting the network parameters, learning rate and

Regularization strategy, and cross-validation and other techniques are applied to evaluate the generalization ability and predictive performance of the model. 5. Predictive analysis and interpretation of results

Combined with the latest medical evidence and clinical guidelines, the model enables prediction of disease progression, evaluation of treatment response and patient prognosis. Detailed prediction reports are generated, including visual data interpretation to help doctors and patients understand the basis of the model predictions. 6. Interactive user interface and feedback loop

The user interface is designed to be intuitive and easy to use, helping doctors and patients view personalized prediction results and

Input of new clinical information or feedback; a feedback mechanism has been established to improve the user interface and the performance of the model based on

to continuously optimize user usage and feedback. 7. System administration, monitoring and maintenance

Regular assessments of system performance and security are carried out to ensure

to ensure data protection and privacy; data backup and

Disaster recovery plans are implemented to ensure system stability and

To guarantee data security.

The heart of the present invention is an innovative multimodal neural

Network structure capable of processing and integrating various types of medical data and extracting in-depth features for predictive analysis of leukemia.

Furthermore, the multimodal neural network used in the present invention comprises

Network structure has the following main features and components: 1. The input layer:

Designed with multiple input channels that include genomic data, clinical data,

Time series data and HO-1 expression data; each channel prepares the data through special preprocessing steps (normalization and data transformation). 2. Parallel processing subnetworks:

Convolutional subnetwork: for processing genomic data, contains several

Convolutional layers and pooling layers.

Dense subnet: used for clinical data and consists of several fully connected layers.

Sequence processing subnetwork: uses a recurrent LSTM neural network to

Processing time series data.

HO-1 expression subnetwork: another dense subnetwork for processing HO-1

Expression data. 3. The level of feature fusion:

Connects and integrates features from different subnetworks; this layer serves as

Convergence point for the results of the subnetworks to ensure comprehensive fusion of the information/ 206274. 4. Multi-Head Attention Mechanism Layer:

The Multi-Head Attention Mechanism layer is the central link for the processing and analysis of data; this layer is implemented as follows:

A. Scaled Dot Product Attention: This layer implements the scaled dot product attention mechanism, which is an effective method to handle complex relationships between different features; in this mechanism, each

Attention “head” attention scores for different parts of the input data and is thus able to detect different patterns and relationships in the data.

B. Multi-head structure: Eight attention heads are set up in the network, each of which focuses on a different aspect of the data; this multi-head structure enables the network to analyze the data from multiple perspectives simultaneously, which

Improved understanding of the data.

C. Attention layer outputs: The outputs of each head are first calculated independently, and then these outputs are integrated together in the

Rule by splicing or weighted summation; the final outputs are these integrated attention signals that combine information from different minds and are used to capture complex relationships in multimodal data. 5. Deep feature extraction layer:

Comprises several fully connected layers, each followed by a ReLU

Activation function; residual connections in the entire feature extraction layer to

To alleviate the training problem of the deep network; batch normalization is used after each fully connected layer to optimize the training process and improve model stability. 6. The output layer:

The output layer is directly connected to the deep feature extraction layer to convert the learned high-level features into specific prediction results; a

A variety of output nodes are designed to correspond to different prediction tasks, such as disease progression, response to treatment and patient prognosis; depending on the different prediction tasks, the output layer uses either a

Sigmoid or softmax activation function.

If the HO-1 expression values are used to determine the specific stage of leukemia (e.g. early, middle, late), a softmax

Activation function suitable.

If HO-1 expression levels are used to predict a binary outcome of leukemia (e.g., whether the disease will return), a sigmoid

activation function can be used.

If HO-1 expression levels are used to predict response to a particular

treatment (e.g. good or poor response), preferably a

Sigmoid function suitable.

If HO-1 expression levels are used to predict response to a particular

treatment, a softmax function should preferably be used if there is more than one possibility of response to treatment (e.g. complete

response, partial response, no response).

Description of the attached drawings LU506274

Figure 1 shows a diagram of the system architecture of the present invention;

Figure 2 shows a diagram of the multimodal neural network architecture used in the present system.

Detailed description

In this embodiment, the features and applications of the architecture and the

Components of a system for monitoring the disease progression of leukemia on the

The basis of modulation of HO-1 expression is described in detail. The system uses a multimodal deep learning technique that combines data such as genomic data, clinical data,

Laboratory test results and HO-1 expression levels combined to predict disease progression,

To predict and evaluate treatment response and patient prognosis in acute myeloid leukemia (AML).

In connection with Figure 1, the system architecture of the present invention comprises the following main modules: 1. Data interface module: Direct interfaces to hospital information systems,

Gene sequencing platforms and laboratory equipment to analyze patient records, genomic data,

laboratory test results, etc. in real time; supports a variety of medical

data formats to ensure data integrity and accuracy. 2. Data processing and integration module: pre-processing of collected data, including data cleaning (removing errors and missing values), conversion (standardization of data formats) and integration (unification of information from different data sources); special attention is paid to the fusion of multimodal data to ensure effective integration of the data into the model. 3. Centralized data warehouse: designed as a high-performance data storage solution to ensure secure and fast data access; supports big data technologies to facilitate the processing and analysis of large data sets. 4. Advanced feature engineering module: performs in-depth data analysis to extract key features relevant to the diagnosis and treatment of leukemia, with special emphasis on automatic feature learning capabilities. 5. Multimodal deep learning module: building neural network models with multiple

Attention mechanisms and complex feature extraction layers that are responsible for

characteristics of leukemia are trained and optimized. 6. Decision support system: Providing data-driven

Decision support for doctors and patients using the results of the deep learning

model, including disease prediction, treatment recommendation and prognosis evaluation. 7. Interactive user interface: Providing an intuitive user interface for

Doctors and patients to query information, view prediction results and provide feedback. 8. Continuous learning and optimization module: continuously optimize the performance and accuracy of the model based on user feedback and newly collected data. 9. System management and monitoring module: responsible for daily operation and maintenance, including performance monitoring, safety management and

Data backup.

Furthermore, the workflow of the present invention is as follows: 1. Data acquisition and reception

The system has interfaces to several data sources, such as electronic 906274

Medical records in hospitals and gene sequencing laboratories to receive detailed patient records, genetic test results, blood analysis data and specific biomarker data, including HO-1 expression levels, in real time.

This data can be divided into four broad categories:

A. Genomic data: sequencing data of the patient's entire genome, including gene variants, gene expression patterns, etc.; targeted sequencing results for specific genetic markers, e.g. for certain genes associated with AML.

B. Clinical data: patient medical records, including diagnostic information,

Medical history, results of physical examination, treatment records, including

Drug regimen, dosage and response; laboratory test results, such as hematological

Tests, biochemical indicators, etc.

C. Time series data: Patient follow-up data, including clinical

Indicators for regular examinations, response to treatment and

Records of symptoms; physiological measurements over time, such as blood pressure,

blood sugar etc.

D. HO-1 expression data: Expression levels of the HO-1 protein and gene expressed in the

Usually determined by blood tests or tissue biopsies; correlation data between HO-1

Expression and disease progression as well as response to therapy. 2. Data preprocessing and quality control

A. Data cleansing: identification and correction of erroneous data, treatment of

Outliers and filling missing values. To ensure data integrity, advanced algorithms are used to estimate and interpolate missing data.

B. Data standardization and normalization: Conversion of data from different

sources into a uniform format and scope to reduce the impact of

Reduce variability of data sources.

C. Data validation: Quality control of data to ensure the accuracy and reliability of the input data. 3. Feature engineering and data integration

A. Methods for feature extraction: For genomic data, the

Principal component analysis (PCA) and gene expression network analysis are used to identify important gene variants and expression patterns; for clinical data,

Time series analysis and trend analysis algorithms are used to identify changes in clinical

metrics; and for drug response data, drug response network modeling and pharmacodynamic modeling are used to predict patient response to specific

to evaluate and predict medications.

B. Data integration techniques: Structured data (e.g., laboratory test results) are integrated using standardized processing (e.g., unit conversion, range adjustment); unstructured data (e.g., physician notes, imaging reports) are integrated using

Natural language processing (NLP) techniques such as named entity recognition and

Sentiment analysis converted into structured formats; data from various sources and

Formats are effectively integrated through collaborative filtering and hybrid models.

C. Feature selection and optimization: Machine learning algorithms such as Random

Forest and Gradient Boosting Decision Tree are applied to select the most informative features based on the model's feature importance rating/506274

Correlation analyses are performed on the feature set to reduce redundant features, and the feature set is further optimized by L2 regularization methods.

Through the methods mentioned above, the system is able to identify the most valuable

Extract information from the rich and complex multimodal data and integrate it into a comprehensive picture of the state that provides high-quality inputs for deep

Learning model delivers. 4. Training and optimization of the deep learning model

The Multi-Head Attention mechanism allows the model to focus on the most important

information in the data and improves the ability to capture complex patterns; the model is tuned using a validation set, including the

Adjustment of the network parameters, the learning rate and the regularization strategy; and

Cross-validation and other techniques are used to assess the generalization ability and

To evaluate the predictive performance of the model. 5. Predictive analyses and interpretation of results

Combined with the latest medical knowledge base and clinical guidelines, the model provides predictions of disease progression, treatment response assessment and patient prognosis; detailed prediction reports are generated, including visual data interpretation, to help doctors and patients understand the basis of

Understanding model predictions. 6. Interactive user interface and feedback loop

The user interface is designed to be intuitive and easy to use and

Helped doctors and patients view personalized prediction results and enter new clinical information or feedback; a feedback mechanism was established to improve the user interface and model performance based on

to continuously optimize user usage and feedback. 7. System administration, monitoring and maintenance

Regular assessments of system performance and security are carried out to ensure

to ensure data protection and privacy; data backup and

Disaster recovery plans are implemented to ensure system stability and

To guarantee data security.

The heart of the present invention is an innovative multimodal neural

Network structure capable of processing and integrating various types of medical data and extracting in-depth features for predictive analysis of leukemia.

In connection with Figure 2, the multimodal neural network structure used in the present invention comprises the following main features and components: 1. The input layer:

Designed with multiple input channels that include genomic data, clinical data,

Time series data and HO-1 expression data; each channel prepares the data through special preprocessing steps (normalization and data transformation). 2. Parallel processing subnetworks:

Convolutional subnetwork: for processing genomic data, contains several

Convolutional layers and pooling layers.

Dense subnet: used for clinical data and consists of several fully connected layers. LUS06274

Sequence processing subnetwork: uses a recurrent LSTM neural network to

Processing time series data.

HO-1 expression subnetwork: another dense subnetwork for processing HO-1

Expression data. 3. The level of feature fusion:

Connects and integrates features from different subnetworks; this layer serves as

Convergence point for the results of the subnetworks to ensure comprehensive fusion of information. 4. Multi-Head Attention Mechanism Layer:

The Multi-Head Attention Mechanism layer is the central link for the

Processing and analysis of data; this layer is implemented as follows:

A. Scaled Dot Product Attention: This layer implements the scaled dot product attention mechanism, which is an effective method to handle complex relationships between different features; in this mechanism, each

Attention, head attention scores for different parts of the input data and is thus able to detect different patterns and relationships in the data.

B. Multi-head structure: Eight attention “heads” are set up in the network, each focusing on a different aspect of the data; this multi-head structure allows the network to analyze the data from multiple perspectives simultaneously, which

Improved understanding of the data.

C. Attention Layer Outputs: The outputs of each head are first calculated independently, and then these outputs are integrated together, usually by

Splicing or weighted summation; the final outputs are these integrated

Attention signals that combine information from different heads and

Capturing complex relationships in multimodal data. 5. Deep feature extraction layer:

Comprises several fully connected layers, each with a ReLU

Activation function follows; the residual connectivity is determined throughout the

feature extraction layer is used to mitigate the training problems of the deep network; batch normalization is used after each fully connected layer to

to optimize the training process and improve model stability. 6. The output layer:

The output layer is directly connected to the deep feature extraction layer to convert the learned high-level features into specific prediction results; a

A variety of output nodes are designed to correspond to different prediction tasks, such as disease progression, response to treatment and patient prognosis; depending on the different prediction tasks, the output layer uses either a

Sigmoid or softmax activation function.

If the HO-1 expression level is used to determine the specific stage of leukemia (e.g. early, middle, late), a softmax activation function is preferable.

Preferably, a sigmoid activation function should be used if the HO-1

Expression values can be used to predict a binary outcome of leukemia (e.g.

B. whether the disease will recur or not).

If HO-1 expression levels are used to predict response to a particular&/506274

treatment (e.g. good or poor response), preferably a

Sigmoid function can be used.

Is the HO-1 expression level used to predict response to a specific

treatment, a softmax function should preferably be used when there are several possibilities for response to treatment (e.g. complete

response, partial response, no response).

In summary, the present invention proposes a system for monitoring the

disease progression of leukemia based on the regulation of HO-1 expression, combining the latest multimodal deep learning technology with an extensive medical

data analysis, with the aim of providing comprehensive, efficient and accurate support for the

diagnosis, treatment and prognosis of acute myeloid leukemia (AML). Through the comprehensive use of genomic data, clinical data, laboratory test results and HO-1

expression levels, this system is able to analyze the complexity of the disease in depth and provide doctors and patients with a more individual and accurate basis for

to provide treatment decisions.

The specific embodiments described here represent only one possible

implementation of the system and are intended to show how an effective system for monitoring the

leukemia progression can be constructed according to the principles and concepts of the present invention. It should be understood that the present invention is not limited to the specific embodiments disclosed herein, but is applicable to various

may be subject to deformation, substitution and modification without affecting the spirit and

scope of the present invention is exceeded.

Claims (4)

Claims LU506274 1. A system for monitoring disease progression in leukemia based on the regulation of HO-1 expression, characterized in that the system comprises: A data interface module: configured to interface with a hospital information system, a gene sequencing platform and a laboratory device for real-time collection of a patient's medical records, genomic data, laboratory test results and HO-1 expression values; Data processing and integration module: for sorting, standardizing and integrating the collected data, in particular fusion of multimodal data; Central data warehouse: for storing the processed data sets; Advanced feature engineering module: for extracting key features relevant to leukemia diagnosis and treatment from the processed data set; Multimodal deep learning module: It contains neural network models with multiple attention mechanisms and complex feature extraction layers for analyzing the extracted features in connection with predicting disease progression, response to treatment and assessing patient prognosis; Decision support system: used for outputting the neural network model to provide treatment recommendations and prognostic assessments; Interactive user interface: used to present prediction results and collect user feedback; Continuous learning and optimization module: used to continuously optimize the performance and accuracy of the model based on user feedback and newly collected data; System management and monitoring module: responsible for daily operation and maintenance, including performance monitoring, security management and data backup. 2. A system for monitoring leukemia progression according to claim 1, characterized in that the neural network structure of the multimodal deep learning module is specifically designed for processing multimodal data in leukemia and comprises parallel processing subnetworks tailored to different types of data sources. 3. A leukemia progression monitoring system according to claim 1 or 2, characterized in that the multi-head attention mechanism of the multimodal deep learning module comprises eight independent attention heads for parallel processing and analysis of data to improve the understanding of complex relationships between different data features and to increase the accuracy and efficiency of predictions. 4. A leukemia progression monitoring system according to any preceding claim, capable of performing multiple types of disease predictions, including disease progression, response to treatment, and patient prognosis, using a deep learning model to analyze multimodal data to provide important information.