CN-121999978-A - Customized machine learning training for radiation therapy clinics

Abstract

Embodiments of the present disclosure relate to customized machine learning training for radiation therapy clinics. Disclosed herein are methods for selecting and preparing patient data to facilitate adoption and custom training of machine learning models in a clinical setting, particularly for radiotherapy treatment planning. The disclosed embodiments simplify the customized training process by an automated workflow that includes pre-filtering patient metadata, retrieving related DICOM files, optional data anonymization, and generating training data. The data is then organized into a format suitable for machine learning training. The embodiments discussed herein reduce manual labor, minimize mistakes, and speed up the integration of machine learning into a clinical workflow, enabling clinics to train and implement predictive models that replicate specific clinical practices, thereby increasing treatment accuracy and improving patient outcome.

Inventors

• M. Hacala
• E. Chezler
• LAAKSONEN HEIKKI ANTERO

Assignees

• 西门子医疗国际股份有限公司

Dates

Publication Date

20260508

Application Date

20251105

Priority Date

20241107

Claims (20)

1. 1. A method for training and integrating a machine-learned ML model for radiation therapy treatment planning, the method comprising: receiving, by at least one processor, one or more radiation therapy treatment attributes for training the ML model; Retrieving, by the at least one processor, patient data from a database accessible to a clinic, the database storing treatment data associated with a previously treated patient set, the patient data corresponding to a subset of the patient set satisfying the one or more radiation therapy treatment attributes; Concatenating, by the at least one processor, the retrieved patient data with one or more of a medical file or other medical file and corresponding metadata associated with previously performed treatments for each patient within the subset of the patient set; Generating, by the at least one processor, a training data set based on the patient data and the one or more digital imaging and communication and metadata in the cascaded medical file or other medical file by changing at least a file structure of the training data set according to a configuration file of the ML model, and Training, by the at least one processor, the ML model using the training data set such that the ML model is customized to the clinic.
2. 2. The method of claim 1, wherein the machine learning model is trained using only the training data set.
3. 3. The method of claim 1, wherein the machine learning model was previously trained using a secondary training dataset and is trimmed to the clinic.
4. 4. The method of claim 1, further comprising: At least one of the digital imaging and communication or the metadata in the patient data, cascaded medical files or other medical files is anonymized by the at least one processor.
5. 5. The method of claim 1, further comprising: the training data sets are clustered, by the at least one processor, into a plurality of consistent subsets.
6. 6. The method of claim 1, further comprising: when an outlier data point is identified, the outlier data point within the training data set is removed by the at least one processor.
7. 7. The method of claim 1, wherein the radiation therapy treatment attributes correspond to a particular treatment technique.
8. 8. The method of claim 1, further comprising: The training data set is deduplicated by the at least one processor by removing data associated with patients meeting a similarity threshold.
9. 9. A computer-readable medium for training and integrating a machine-learned ML model for radiation therapy treatment planning, the computer-readable medium comprising instructions that, when executed, cause a processor to: Receive one or more radiation therapy treatment attributes for training the ML model; Retrieving, by the at least one processor, patient data from a database accessible to a clinic, the database storing treatment data associated with a previously treated patient set, the patient data corresponding to a subset of the patient set satisfying the one or more radiation therapy treatment attributes; Concatenating and communicating the retrieved patient data with one or more of a medical file or other medical file and corresponding metadata associated with previously performed treatments for each patient within the subset of the patient set; Generating a training data set based on the patient data and the one or more digital imaging and communication and metadata in the cascaded medical file or other medical file by changing at least a file structure of the training data set according to a configuration file of the ML model, and The ML model is trained using the training dataset such that the ML model is customized to the clinic.
10. 10. The computer-readable medium of claim 9, wherein the machine learning model is trained using only the training data set.
11. 11. The computer-readable medium of claim 9, wherein the machine learning model was previously trained using a secondary training dataset and is trimmed to the clinic.
12. 12. The computer-readable medium of claim 9, wherein the instructions further cause the processor to anonymize at least one of the digital imaging and communication or the metadata in the patient data, cascaded medical file, or other medical file.
13. 13. The computer-readable medium of claim 9, wherein the instructions further cause the processor to cluster the training data set into a plurality of homogeneous subsets.
14. 14. The computer-readable medium of claim 9, wherein the instructions further cause the processor to remove the outlier data point within the training data set when the outlier data point is identified.
15. 15. The computer readable medium of claim 9, wherein the radiation therapy treatment attribute corresponds to a particular treatment technique.
16. 16. The computer-readable medium of claim 9, wherein the instructions further cause the processor to deduplicate the training data set by removing data associated with patients meeting a similarity threshold.
17. 17. A computer system for training and integrating a machine-learned ML model for radiation therapy treatment planning, the computer system comprising a processor configured to: Receive one or more radiation therapy treatment attributes for training the ML model; Retrieving, by the at least one processor, patient data from a database accessible to a clinic, the database storing treatment data associated with a previously treated patient set, the patient data corresponding to a subset of the patient set satisfying the one or more radiation therapy treatment attributes; Concatenating and communicating the retrieved patient data with one or more of a medical file or other medical file and corresponding metadata associated with previously performed treatments for each patient within the subset of the patient set; Generating a training data set based on the patient data and the one or more digital imaging and communication and metadata in the cascaded medical file or other medical file by changing at least a file structure of the training data set according to a configuration file of the ML model, and The ML model is trained using the training dataset such that the ML model is customized to the clinic.
18. 18. The computer system of claim 17, wherein the machine learning model is trained using only the training data set.
19. 19. The computer system of claim 17, wherein the machine learning model was previously trained using a secondary training dataset and is trimmed to the clinic.
20. 20. The computer system of claim 17, wherein the processor is further configured to anonymize at least one of the digital imaging and communication or the metadata in the patient data, cascaded medical file, or other medical file.

Description

Customized machine learning training for radiation therapy clinics Technical Field The present application relates generally to office-specific radiation therapy planning systems, and in particular to the customization of a trained machine learning model for radiation therapy planning to improve the operational efficiency thereof. Background Radiation Therapy Treatment Planning (RTTP) is a complex process that includes specific guidelines, protocols, and instructions employed by different medical professionals, such as clinicians, medical device manufacturers, therapists, etc. Due to the extreme nature of the radiation emitted by the radiotherapy machine, all instructions must be precisely followed. The field geometry as used in the context of RTTP refers to various attributes or settings of the radiation therapy machine when the patient receives a prescribed radiation therapy dose. For example, the prescribing physician may identify the structure (e.g., the patient's organ to be treated or the tumor to be eradicated) and the corresponding dose. In addition, other parties (e.g., clinicians or machine manufacturers) may determine the positioning attributes (e.g., angles) of the gantry and patient on the treatment couch to provide optimal treatment. To increase the efficiency of this process, many clinics use their own machine learning models that are trained using a central entity. For example, an entity may provide a machine learning model to a particular clinic, where the clinics may customize or fine tune the machine learning model using their own patient data and/or clinic-specific rules and protocols. Such customized machine learning methods may be integrated into a plan optimizer platform, where machine learning models trained using patient groups are fine-tuned for a particular clinic. The primary purpose of these methods is to assist medical professionals and dosimeters in creating high quality, consistent treatment plans for cancer patients by utilizing past treatment data using custom models for specific clinics. Using this paradigm, each clinic can train its model (or adapt by further training a pre-trained model) using its own historical treatment data or any customized data set. This allows the model to reflect specific clinical practices and preferences of the clinic, ensuring that the generated treatment plan is consistent with the criteria of its clinic. However, clinical specific radiation therapy planning approaches face some challenges. For example, the system is labor intensive and error prone as it involves each clinic preparing data for its model training. For example, a clinician must manually identify and select relevant patient data from a large complex database to train a clinic-specific model. This process includes extracting the treatment plan, ensuring that the data is properly marked, and organizing it into a format that can be used to train the model while ensuring the quality of the data. Such manual work is time consuming, and requires a high level of expertise, making it a significant challenge for medical professionals. Furthermore, the quality of the model will depend directly on the subjective expertise of the medical professional at each clinic, which is undesirable and inconsistent. Finally, because models are sensitive to variability in the data used for training, they may not operate effectively. For example, some models require a homogenous data set to function effectively, meaning that any inconsistencies or outliers (outliers) in the data can negatively impact the performance of the model. This further complicates the data preparation process as it requires careful selection and management of the data to ensure that it meets the necessary criteria for training. Disclosure of Invention For the above reasons, there is a need for a system that can adapt or otherwise customize a computer model (e.g., AI or machine learning or more conventional model) for a particular clinic. Using the methods and systems discussed herein may allow for faster and more efficient training times, and sometimes use less computing resources. Furthermore, machine learning models trained using the methods and systems discussed herein may be customized to a clinic so that their predictions are more accurate. Accordingly, the methods and systems discussed herein provide functional and technical improvements specific to the machine learning field. The methods and systems discussed herein address challenges associated with manual data processing in training of machine learning models for radiation therapy treatment planning. Currently, clinics must manually sort patient data from complex databases, a process that is both time consuming and error prone. This manual approach can lead to inconsistencies and inefficiencies because the clinic needs to extract relevant data, organize it, and ensure its quality for machine learning model training. The methods and systems discussed herein provide an autom