A process for automatic generation of structured medical report for cancer imaging

Abstract

This invention discloses a process characterized by two parts, a resident part and a cloud-based part, each part is composed of set of modules. The invention is deployed through four phases: the resident part installation phase, the configuration of connection between resident part and medical institute database phase, the deployment of artificial intelligence models phase, and the configuration of the access from the resident part to the cloud-based part phase. Together these parts will provide the automatic generation of standardized medical report for cancer imaging in patients with cancer. This invention can be used to increase quality of patient care, optimize efficiency in radiology clinics and within cancer care team, and optimize resources allocated to health system.

Description

A PROCESS FOR AUTOMATIC GENERATION OF STRUCTURED MEDICAL REPORT FOR CANCER IMAGINGDESCRIPTIONFIELD OF INVENTION

[0001] This invention discloses a system characterized by two parts, a resident part and a cloud-based part, each part is composed of set of modules. The system is deployed through four phases: the resident part installation phase, establishment of the connection from the resident part to medical database for automatic data import, deployment of artificial intelligence models to analyze data and make predictions, and the configuration of the access from the resident part to the cloud-based part phase. Together the system will provide the automatic generation of structured radiology report (medical report) for cancer imaging.BACKGROUND ART

[0002] Radiological and nuclear scans, for example computer tomography (CT), magnetic resonance imaging (MRI), positron emission tomography I computer tomography (PET / CT), single-photon emission computed tomography (SPECT), single-photon emission computed tomography / computer tomography (SPECT / CT), or scintigraphy are imaging procedures used in cancer imaging to help locate, diagnose, and stage cancer. Radiological and nuclear imaging scans performed in patients with cancer are often referred to as cancer imaging.

[0003] Physicians that review, interpret, and report radiological and nuclear scans include diagnostic radiologists and nuclear medicine physicians (henceforth referred together to as radiologists).

[0004] Cancer care team is the interdisciplinary team involved in care of patients with cancer and includes medical oncologists, surgical oncologists, pathologists, radiationoncologists, radiologists, and primary care physicians, among other specialists and health workers. Clinical decision making in oncology involves the consideration of a complex set of information about patient clinical history, cancer history and biology, results of cancer imaging, and patient prognosis and preferences. Decision making in cancer patients is at the intersection of evidence-based medicine and the art of shared decision making between patients and their clinicians.

[0005] Cancer imaging plays a critical role in management of patients with cancer since the results of cancer imaging are usually the last, and often most important, piece of information required by cancer care team to make a clinical decision for patient. Ideally, the content of a cancer imaging report provides information on patient clinical history, pathology results, information about the process of images acquisition, cancer stage, cancer tumor description, biologic characteristics, patient prognosis based on imaging findings and clinical history, and therapy response evaluation, among other details. Such comprehensive cancer imaging report can be a one-stop shop for cancer care team since it can include all information needed for clinical decision-making process. The cancer imaging report ultimately directs patient care, serves as legal documentation of services provided and possible technical limitations, and serves as a communication tool between physician and patient.

[0006] We describe below several current issues from day-to-day practice of radiologists that often leads to unclear or even incomplete medical reports for cancer imaging:

[0007] 1 . The radiological report is often free-text and its structure, style, wording, and layout differ among hospitals and from radiologist to radiologist. Text-only qualitative reports are criticized because of numerous inconsistencies in their content, lack of structure and diminished comprehensibility of relevant information.

[0008] 2. When writing the report, the radiologist typically reviews patient’s Electronic Health Records (EHR), extract relevant clinical information, and describe them in the report. This task is ineffective, time consuming, often incomplete, and prone to human error.

[0009] 3. When writing the report, the radiologist typically reviews in the Radiology Information System (RIS) technical information about the process of images acquisition and describes them in the report. This task is ineffective, time consuming, often incomplete, and prone to human error.

[0010] 4. The primary tumor, nodes, metastasis (TNM) classification of the American Joint Committee on Cancer / Union for International Cancer Control is a widely accepted and standardized staging system for cancer. Nevertheless, only 25% of radiologists routinely report cancer stage in radiology reports and 58% of radiologists do not report cancer stage in the radiological report because they are unable to remember the staging classification (PMID: 27032970). These data highlight the need for optimization of radiology reports and developing systems to improve reporting workflow.

[0011] 5. 54% of radiologists are affected by burnout according to the Medscape’s 2023 Physician Burnout & Depression Report and the leading factors contributing to burnout is excessive workload and bureaucratic tasks. Time-consuming and ineffective bureaucratic tasks are related to creating the radiological report, e.g. describing in the report information that is already available in institution medical system and is not directly related to medical act of reporting the results of medical image interpretation.

[0012] Due to above mentioned limitations, the medical reports for cancer imaging are often brief providing strictly a qualitative description of imaging findings about cancer tumor. Such reports are often difficult to interpret in a clinical context by cancer care team and leave the clinician in a clinical dilemma. In such situations, the clinician access medical images to be able to interpret the report in the clinical context. During clinical decision-making process, cancer care teams need to access multiple information about patient and cancer history stored in different locations of patient electronical health records or radiology information system. This information includes patient clinical history, laboratory test results, pathology report, radiology report, and radiology images. This process is complex, time-consuming, inefficient, and is prone to human error.

[0013] Below we provide some new findings in this area that also evidence the lack of systems to support radiologists create a comprehensive cancer imaging report that supports cancer care team and public health system.

[0014] EP4176377A1 describes artificial intelligence (Al)-based systems and methods that improve cancer detection and characterization of cancer lesions via automated analysis of radiological images. The proposed Al-based systems for cancer detection, segmentation, and classification provide a basis for further characterization of cancer lesions (e.g. total tumor burden), and estimation of cancer severity and patient risk. However, it does not describe the process of generation of a structured radiological report for cancer imaging.

[0015] US 2022 / 0336071 A1 describes a system for reporting radiological medical images and the system comprises: means for receiving one or more medical images, means for receiving from a repository and / or creating a machine-readable structured report based on information extracted from analysis of medical images. The system can automatically incorporate into the radiological report the section related to cancer findings on medical images. However, it lacks integration in the report of other sections required for a comprehensive medical report for cancer imaging, e.g. patient information, cancer clinical history, technical information of the process of image acquisition, patient prognosis.

[0016] US10445462B2 describes an artificial intelligence findings system including a findings engine that receives medical image data to generate findings based on the medical image data and image interpretation algorithms. A client viewer presents the findings to the user and an adjustment engine allows the user to adjust the findings to produce a report. However, it lacks integration in the report of other sections required for a complete medical report for cancer imaging, i.e. patient information, cancer clinical history, technical information of the process of image acquisition.

[0017] US 2019 / 0074074 A1 describes systems comprising of displaying a medical image of a region of interest, determination of an image characteristic of the regionof interest, generation of a recommendation based on the image characteristic, and generation of an imaging report including recommendation for the patient. However, this invention lacks generation of a complete radiology report for cancer imaging and does not allow the user to consider in the recommendation clinical factors pertinent for clinical management of patients with cancer.GENERAL DESCRIPTION

[0018] The growth of the population and of the number of cancer cases clearly require a new technology capable of satisfying the requirements pointed out above.

[0019] To overcome this gap in today’s technology, this invention reveals a process implemented through a computer program that can:

[0020] i) Analyze medical images of cancer patients and extract relevant information in a standardized way (e.g., cancer stage according to TNM classification),

[0021] ii) Extract relevant clinical information from patient electronic health records and technical information from radiology information system, and,

[0022] ill) Incorporate this information altogether into a structured and comprehensive report. This will increase the quality of medical report of cancer imaging and ultimately the efficiency of cancer care teams. Some reasons are:

[0023] a. First, it would decrease the time spent by cancer care teams in interpreting results of cancer imaging and in clinical decision-making process of patients with cancer. This will allow clinical cancer teams to spend more time with patients and avoid medical errors due to increased workload and burnout among oncologists.

[0024] b. Second, such a process will decrease the time spent by radiologists to report results of cancer imaging, and by doing that it will avoid medical errors as it will decrease the workload and reduce burnout among radiologists.

[0025] Overall, such a process will improve quality of health care for patients with cancer and even allow for new developments in the field of technical analysis of cancer images, in particular comparative studies of images, diagnostics, and treatments (figure1).

[0026] Therefore, it is critically important to the quality of healthcare provided to have an improved system for automatically generating a comprehensive cancer imaging report that incorporates:

[0027] • Results of medical images of cancer,

[0028] • Technical information of the process of image acquisition,

[0029] • Clinical information about patient and cancer history,

[0030] • Patient prognosis to strategize patient clinical management and patient response to cancer treatment.

[0031] • And the recorded cancer information, the more complete and more structured the better, into an electronic data repository.

[0032] This is clearly a gap in the prior art. To address this problem this invention reveals a process consisting of two parts, each part with its own installation process. The resulting process for automatic generation of structured cancer imaging report creates a process that converges to provide more comprehensive and accurate, better structured, and clear reports for specialists involved in cancer care and patients.

[0033] We start by describing the parts, the Resident Part and the Cloud Part, and then we explain the deployment and functioning of the invention.

[0034] The Resident Part is composed of a set or subset of 10 modules (figure 2).

[0035] Input Data Interface: Module responsible for the input of patient data at user interface. This module also makes a quick check for data format and structure before passing it on to the CPU and the destination database according to the kind of data being entered.

[0036] CPU: This module is responsible for the processing of data in the resident part of the system. All data from all modules are received, transformed, and redirected through this module.

[0037] User Data: all patient data is stored and classified in this database.

[0038] Security: This module is where the data security and system security at residentpart level are implemented. It is responsible for data encryption and other cyberdefense strategies to guarantee privacy and accuracy of data and system.

[0039] Medical Report: This module is responsible for receiving information from several modules, for integrating them in the pertinent sections, and for automatically generating a standardized medical report.

[0040] Input Clinical Data: Module responsible for input patient clinical data from institution electronic health record (EHR) databases. Clinical data include information of patient demographics, clinical history, cancer history, laboratory tests, genomics and epigenomics. The import process is automatic using application programming interface (API) systems. Data recognition and extraction can be done using natural language processing and machine learning techniques, depending on the type of data being extracted. If information required for medical report cannot be imported automatically by this module, it is signaled, and the missing information is introduced manually by a radiologist using the Input Data Interface module. This module also makes a quick check for data format and structure before passing it on to the CPU module and the destination database according to the kind of data being entered.

[0041] Input DICOM Data: This module is responsible for inputting technical information that describes the process of medical image acquisition. The module extracts information from metadata of Digital Imaging and Communications in Medicine (DICOM) files. Sets of DICOM files are accessed from the institution Picture Archiving and Communication System (PACS) within the Radiology Information System (RIS) environment. Information is extracted automatically from DICOM metadata and include the type of scanner used for image acquisition, the type and amount of radiopharmaceutical injected, the type and amount of contrast agent injected, time of the scanning procedure, and the location of the scanning procedure, or whether a diuretic was administered for the imaging procedure. The Input DICOM Data module also makes a quick check for DICOM file format and DICOM extracted information that needs to be processed by the CPU module.

[0042] Artificial Intelligence (A. I.) Models: In this module, the pre-programmed A. I. models are run at resident part level and their results are incorporated into model strategies like detection and classification of cancer lesions on medical images. The A. I. models use as input patient DICOM files accessed from institution PACS system and processed by CPU. Systems based on deep learning and machine learning are implemented at this module. Depending on the implementation, this module may be in connection with its cloud-based counterpart for updates and upgrades.

[0043] Predictions: This module is responsible for receiving, transforming, and redirecting the results of A. I. models and for making ex-post-analysis verification. Predictions generated by A. I. models are cancer TNM stage, automatic detection and segmentation of cancer lesions, tumor characteristics (heterogeneity, degree expression of tumor markers such PSMA or FDG etc.), automatic calculation of cancer prognosis based on imaging findings and clinical history, automatic prediction of response to cancer therapy, and / or automatic evaluation of efficacy of cancer therapy.

[0044] Communication: This module manages communications and protocols for assuring that the right data gets to the right place at the right time. This module is also responsible for communication of data between resident part and public health system and is also responsible for assuring that the data from the right resident part gets to the right party at the right time.

[0045] The Cloud part is composed of set of subsets of 9 modules (figure 3):

[0046] Cloud Input: Patient data from each resident part is received at this module. Through this module the system operator that owns the Cloud Part can enter new data and pre-programmed A.I. algorithms into the system. It can also be used to select and clean non-desirable data (bad data) detected by the system, manage the client database, financials, and other management, updating and upgrading activities.

[0047] Cloud Communication: This cloud module manages communications and protocols for assuring that the data from the right resident part gets to the right place at the right time. The module is also responsible for checking communication with residentparts and updating communication protocols whenever it is needed.

[0048] CPU: This module is responsible for the processing of data in the cloud part of the system. All data from all cloud modules are received, transformed, and redirected through this module.

[0049] Cloud Data: All patient data received from resident parts at cloud databases through the Cloud Input module are stored and classified in this database. The Cloud Data module will operate under the most stringent rules of privacy and respect to personal data.

[0050] Cloud Security: This module is where the data security and system security are implemented at cloud level and as whole system, signaling for part levels the need of upgrades / updates. It is responsible for data encryption and other cyber-defense strategies to guarantee privacy and accuracy of data and system.

[0051] Data Analysis: This module performs statistical analyses of all patient data collected at cloud level. Results of analyses can be used, for example, to understand at a global level the type of scanning procedures used and how they differ among institutions or to compare the results of cancer imaging (e.g., cancer stage) among private practice centers versus academic institutions. Results of data analysis can also generate ideas on new A.I. models that can be developed to improve the care of patients with cancer.

[0052] Artificial Intelligence (A. I.): In this module, the pre-programmed A.I. models deployed at resident part level can be updated using patient data collected from multiple resident parts. New A.I. models can also be developed at cloud level to develop new strategies to improve the care of patients with cancer. New or updated A.I. models are deployed at resident part level via Communication modules.

[0053] Predictions: This module is responsible for receiving, transforming, and redirecting the results of A.I. models at cloud level and for making ex-post-analysis verification.

[0054] Cloud Output: This module performs validation on all results and products generated by the Cloud Part and takes care of the distributions of these results to theright interface, be it one or more resident parts or a direct operator interface at cloud level.

[0055] From the viewpoint of functionally, the invention works in four phases.

[0056] For completeness, we assume that the resident and cloud part are already installed, and the client has agreed to provide data to the cloud part which is also integrated with large cancer electronic repositories as referred above.

[0057] Phase 1 : Deployment of the first 6 modules of the Resident Part (subset of the Resident Part)

[0058] In this phase the resident part is installed with 6 modules (CPU, Input Data Interface, Medical Report, User Data, Security, and Communication). The user can entry patient data manually through the Input Data Interface module. All patient data is stored in the User Data module. After data is processed through CPU module, a standardized medical report is generated automatically to provide image interpretation. The medical report can integrate the following sections:

[0059] -> Clinical History: Relevant information about patients and cancer history is provided. It includes patient history (e.g., age, comorbidities), cancer history (e.g., initial diagnosis details, pathology results, cancer stage, treatment history), clinical indication for the scan (e.g., initial staging or re-staging of cancer), and / or laboratory test results (e.g., tumor markers, complete blood count).

[0060] -> Technical information: Include information about the process of image acquisition. It includes the type of imaging technique (e.g., CT, MRI, PET / CT, SPECT, SPECT / CT, or scintigraphy), scanner (type, model, location), contrast agent (type, quantity, time of injection), and / or radiopharmaceutical used (type, quantity, time of injection).

[0061] -> Oncological findings: A summary of cancer stage is provided in a standardized way following TNM classification system.

[0062] -> Descriptive findings: A detailed description of cancer is given including anatomical localization, size, shape, relation to other anatomical structures.

[0063] -> Treatment response evaluation: Changes in tumor noticed on images by radiologists during or after receiving cancer therapy are described.

[0064] -> Eligibility status for therapy: A recommendation according to established guidelines is made whether the patient is eligible for a certain cancer therapy based on imaging findings.

[0065] -> TNM Stage codeline: A codeline that summarizes cancer stage according to TNM classification system.

[0066] -> Pictorials: A graphical representation of cancer localization in the human body.

[0067] -> Patient Summary: Describes in plain English the summary of imaging results to a nonmedical audience.

[0068] A medical report for cancer imaging structured as described above is depicted in figure 4.

[0069] Phase 2: Connection of the Resident Part with Health Institution database.

[0070] In this phase a connection between Resident Part and client institution database is established and additional 2 modules (subsets of the Resident Part) of resident part are deployed:

[0071] Input Clinical Data: Patient data related to Clinical History section described above can now be imported automatically from institution EHR system.

[0072] Input DICOM Data: Patient data related to Technical Information section described above can now be imported automatically from metadata of DICOM images. DICOM is a standard, internationally accepted format to view, store, retrieve and share medical images. Patient DICOM files are stored in institution Picture Archive and Communication System (PACS) I Radiology Information System (RIS). Patient DICOM files are accessed by resident part from institution PACS / RIS.

[0073] Phase 3: Deployment of A.I. models (subset of Resident or Cloud Part)

[0074] In this phase, 2 additional modules are added i.e., A.I. Models and Predictions. The pre-programmed, already calibrated A.I. models are deployed and connected toinstitution database to access patient data and process them. The A. I. models access patient DICOM files from institution PACS / RIS to analyze medical images and make predictions about cancer (e.g., anatomical localization, stage, tumor burden, response to cancer therapy, eligibility for cancer therapy). A connection to the Resident part is also established to provide predictions of A.I. models via the Predictions module after patient data is analyzed. The predictions are ultimately incorporated automatically in the medical report. Other A.I. models can use predictions obtained from DICOM image analysis together with pre-defined patient clinical data stored at User Data module to make new predictions about patient prognosis and therapy response.

[0075] The resident part is now installed on client premises in its entireness with all 10 modules, the connection with hospital database is established, and the A.I. models are supplied already calibrated.

[0076] Phase 4: Deployment of the Cloud Part

[0077] In this phase, the Resident part is connected to a cloud-based database.

[0078] The communication between the Resident part and the Cloud part is established via Communication modules that are interconnected.

[0079] The Resident Part can be connected to one or more Cloud Parts and these Cloud Parts may be private or public. Residents Parts from different locations may be connected to one or more Cloud Parts and this configuration may allow for teleradiology. Both the Resident Part and the Cloud Part may allow for hierarchical structure where one Master Cloud or Resident Part controls others Cloud or Resident Parts respectively. Security configurations distribute access to information to a person or group of persons that should have access to it.

[0080] The Cloud Part is supported by research and data analysis, normally from at least one research platform or institution. Several different submodules may be produced and added to its modules:

[0081] -> Machine Learning algorithms: normally added to the A.I. module.

[0082] -> Cyber-security strategies and cryptography developments: normally addedto the Security module.

[0083] -> Communication developments: normally added to the Communication module. -> Interface and Usability improvements: normally added to one or more input or output modules.

[0084] -> Business models improvements: based on technology developments, patents and diverse business cooperation strategies, some new improvements may be offered on a pay per improvement basis.

[0085] Brief Description of Drawings

[0086] Figure 1 . shows the current framework for clinical decision making in patients with cancer and the framework proposed by the invention. Note in the current workflow that the cancer care team needs to access multiple medical reports located in different sources within the electronic health records (EHR) and picture archiving and communication system (PACS). In the workflow proposed by this invention, all information required by cancer care team for the clinical decision making is available in the cancer imaging report generated by invention.

[0087] Figure 2. shows a schematic representation of the resident part of this invention. The resident part comprises 10 modules that alltogether can generate automatically a structured report for cancer imaging.

[0088] Figure 3. shows a schematic representation of the cloud part of this invention. The cloud part comprises 9 modules that alltogether can generate automatically a structured report for cancer imaging.

[0089] Figure 4. shows an example of a structured report for cancer imaging produced by the invention.

[0090] Figure 5. shows the invention in its most completed form, with the resident part, cloud part, and outside world communicate and work together to generate automatically a structured report for cancer imaging.

[0091] Figure 6. shows a schematic overview on how the structured radiology reports obtained using the invention can generate automatically cancer records. Cancerrecords from multiple healthcare facilities can be automatically centralized, for example, at public health system to obtain accurate cancer statistics.PREFERRED EMBODIMENTS

[0092] The first preferred embodiment for this invention is its implementation for PSMA-PET / CT imaging in prostate cancer. In this case, cancer stage follows the American Joint Committee on Cancer (AJCC) or PROMISE (PMID: 36935345) TNM classification system, therapy response evaluation follows RECIP criteria (PMID: 37432081 ), patient prognosis and response to PSMA-targeted therapy can be predicted using nomograms (PMID: 34246328). A. I. models perform automatic tumor detection and segmentation on PSMA-PET / CT images and predictions include cancer TNM Stage, total tumor burden, tumor expression of prostate-specific membrane antigen (PSMA), among others. Clinical data that are specific for this preferred embodiment and which can also be imported from patient electronical health record include: cancer history (e.g., time of diagnosis of prostate cancer, serum PSA value at time of diagnosis), laboratory tests (serum prostate-specific antigen [PSA] value), previous therapies include surgery (radical prostatectomy), chemotherapy (e.g., docetaxel, cabazitaxel), hormonal therapy (e.g., ADT, abiraterone, enzalutamide), pathology results (Gleason score at diagnosis or e.g., ATM and TP53 mutations for genetics). Clinical state of prostate cancer is also mentioned (i.e., primary prostate cancer, biochemical recurrence, metastatic / nonmetastatic hormone-sensitive prostate cancer, and metastatic / nonmetastatic castration-resistant prostate cancer). Technical information that are specific for this preferred embodiment which can also be imported from PET / CT DICOM metadata include type of imaging technique (i.e., PET / CT), radiopharmaceutical type (e.g., 68Ga-PSMA-11 , 18F-DCFPyl_, 18F-rhPSMA-7.3), injection time of radiopharmaceutical (approximatively 60 minutes prior to scanning time), among others.

[0093] The second preferred embodiment for this invention is the process built anddeployed in its most complete form as described above with reference to figure 5. The process can be applied for imaging lymphoma, melanoma, cancers of the prostate, breast, lungs, and gastrointestinal tract, among other malignancies.

[0094] A third preferred embodiment is the invention implemented only with resident part completely installed (phase 1 , phase 2, and phase 3) as described above with reference to figure 2. In this case the invention loses the capacity of online updates and of gaining in modeling and information from processes like federated learning strategies that could count on networking gains of scale and scope.

[0095] In a fourth preferred embodiment the invention is implemented only with phase 1 and phase 2 of resident part (subsets of the Resident Part) being deployed. In this setting the patient data can be imported automatically from the medical institute database as described above. In this case the system loses the capacity to perform A.I. based analyses. The radiologist interprets medical images of patients and can use representative graphics of human body anatomy to indicate the anatomic location of cancer lesions. The cancer stage following TNM classification are then generated automatically based on anatomical regions indicated to be affected by cancer. The results are incorporated in the medical report.

[0096] In a fifth preferred embodiment the invention is implemented with only phase 1 of resident part (a subset of resident part) being installed. In this setting the patient data is entered manually by the user and no connection with medical institute database is established.

[0097] In a sixth preferred embodiment the invention is implemented to be used to define global strategy to obtained accurate cancer statistics. In this case it may be used for automatic collection of cancer data at healthcare facilities. Data collected can be centralized by, for example, public healthcare systems to generate accurate statistics of cancer data and optimize the use of healthcare resources (figure 6).

[0098] In a seventh preferred embodiment the invention is used to generate data for research purposes to advance clinical care for patients with cancer. In this case it maybe used for automatic collection of patient data at healthcare facilities. The data can be analyzed by research centers to develop new strategies for cancer diagnosis and management that can be implemented through the Cloud Part and there disseminated to the Resident Parts. By doing that, the invention will decrease costs and resources required for manual data extraction from patient medical records at the same time that it increases overall efficiency of the health system.

[0099] In an eight preferred embodiment the invention is implemented only with Cloud Part. In this setting it may be used for collection and centralization of electronic medical data to perform medical research or develop new A.I. algorithms.

[0100] The ninth preferred embodiment for this invention is its implementation for breast cancer. Imaging techniques used to diagnose and stage breast cancer are mammography, ultrasound, CT, MRI and PET / CT. In this case, cancer stage follows the American Joint Committee on Cancer (AJCC) TNM classification system, therapy response evaluation follows RECIST (PMID: 19097774) and PERCIST (PMID: 26909647) criteria, patient prognosis can be predicted using nomograms (https: / / nomograms.mskcc.org / breast / index.aspx). A.I. models perform automatic tumor detection and segmentation on MRI, FDG-PET / CT and FES-PET / CT images and predictions include cancer TNM Stage, total tumor burden, tumor uptake of fluorodeoxyglucose (FDG) or fluoroestradiol (FES), among others. Clinical data that are specific for this preferred embodiment and which can also be imported from patient electronical health record include: cancer history (e.g., time of diagnosis of breast cancer, initial mammogram and ultrasound results), laboratory tests, previous therapies include surgery (mastectomy), chemotherapy, hormonal therapy, immunotherapy, pathology results (estrogen receptor [ER], progesteron receptor [PR], human epidermal growth factor receptor [HER2] status at diagnosis or e.g., ATM and TP53 mutations for genetics). Clinical state of breast cancer is also mentioned (i.e., primary breast cancer, metastatic / nonmetastatic endocrine-sensitive or endocrine-resistant breast cancer). Technical information that are specific for this preferred embodimentwhich can also be imported from DICOM metadata include type of imaging technique (i.e., mammography, ultrasound, CT, MRI and PET / CT), radiopharmaceutical type (e.g., 18F-FDG, 18F-FES), injection time of radiopharmaceutical (approximatively 60 minutes prior to scanning time), among others.

[0101] The tenth preferred embodiment for this invention is its implementation for FDG-PET / CT imaging in lymphoma. In this case, cancer stage follows Lugano classification (PMID: 25113753), therapy response evaluation follows Deauville score (PMID: 19544140), patient prognosis can be predicted using nomograms (e.g., PMID: 35672070). A.I. models perform automatic tumor detection and segmentation on FDG-PET / CT images and predictions include cancer TNM Stage, total tumor burden, tumor uptake of fluorodeoxyglucose (FDG), among others. Clinical data that are specific for this preferred embodiment and which can also be imported from patient electronical health record include: cancer history (e.g., time of diagnosis of lymphoma, lymphoma type), laboratory tests (e.g., complete blood count), previous therapies include surgery, chemotherapy, radiation therapy, monoclonal antibodies, pathology results (lymphoma type: Hodgkin’s lymphoma, Burkitt lymphoma, follicular lymphoma, diffuse large B-cell lymphoma, mantle cell lymphoma, marginal zone B-cell lymphoma, peripheral T-cell lymphoma, B-cell lymphoma). Technical information that are specific for this preferred embodiment which can also be imported from PET / CT DICOM metadata include type of imaging technique (i.e., CT and PET / CT), radiopharmaceutical type (i.e., 18F-FDG), injection time of radiopharmaceutical (approximatively 60 minutes prior to scanning time), among others.

Claims

CLAIMS1 . A process for automatic generation of structured medical report for cancer imaging characterized by a Resident Part composed of a set or a subset of 10 modules: Input Data Interface, CPU, User Data, Security, Medical Report, Input Clinical Data, Input DICOM Data, Artificial Intelligence (A. I.), Predictions and Communication; and a Cloud Part composed of a set or subset of 9 modules: Cloud Input, Cloud Communication, CPU, Cloud Data, Cloud Security, Data Analysis, Artificial Intelligence (A.I.), Predictions and Cloud Output that are deployed to fulfil the the tasks of data collection of cancer patients, collecting imaging results of cancer patients, processing collected information and generating a report including results of cancer medical images of cancer, technical information of the process of image acquisition including the TNM stage, clinical information about patient and cancer history, patient prognosis to strategize patient clinical management and that can save all recorded cancer information into an electronic data repository and these two parts together produce a medical report for cancer imaging.

2. A process for automatic generation of structured medical report for cancer imaging characterized by a Resident Part composed of sets and subsets of 10 modules: Input Data Interface, CPU, User Data, Security, Medical Report, Input Clinical Data, Input DICOM Data, Artificial Intelligence (A.I.), Predictions and Communication that together produce a medical report for cancer imaging.

3. A process for automatic generation of structured medical report for cancer imaging characterized by a Cloud Part composed of sets and subsets of 9 modules: Cloud Input, Cloud Communication, CPU, Cloud Data, Cloud Security, Data Analysis, Artificial Intelligence (A.I.), Predictions and Cloud Output to fulfil the the tasks of data collection of cancer patients, collecting imaging results of cancer patients processing collected information and generating a report including results of cancer medical images of cancer including the TNM stage, technical information of the process of image acquisition, clinical information about patient and cancer history, patient prognosis to strategizepatient clinical management and that can save all recorded cancer information into an electronic data repository.

4. A process for automatic generation of structured medical report for cancer imaging characterized by a Cloud Part according to claim 3 where the report and the processing, partially or in total, is done in a resident part with data communicated through the communication module.

5. A process for automatic generation of structured medical report for cancer imaging according to claim 1 , 2 or 3 characterized by being configured to an application for PSMA PET / CT imaging in prostate cancer where cancer stage follows AJCC or PROMISE TNM classification system, therapy response evaluation follows PPP and RECIP criteria, patient prognosis is predicted using nomograms, A.I. models perform automatic tumor detection and segmentation on PSMA-PET / CT images, predictions including cancer TNM Stage, total tumor burden, tumor expression of prostate-specific membrane antigen (PSMA), among others, clinical data includes cancer history, laboratory tests, previous therapies and it also includes the clinical state of the prostate cancer and further technical information about the process of PSMA-PET / CT image acquisition, where all the information above will make together a radiology report for PSMA-PET / CT in patients with prostate cancer.

6. A process for automatic generation of structured medical report for cancer imaging according to claim 1 , 2 or 3 characterized by being configured to an application for CT, MRI and PET / CT imaging in breast cancer where cancer stage follows AJCC TNM classification system, therapy response evaluation follows RECIST and PERCIST criteria, patient prognosis is predicted using nomograms, A.I. models perform automatic tumor detection and segmentation on PET / CT images, predictions including cancer TNM Stage, total tumor burden, and tumor uptake, among others, clinical data includes cancer history, laboratory tests, previous therapies and it also includes the clinical state of the breast cancer and further technical information about the process of CT or PET / CT image acquisition, where all the information above will make together aradiology report for CT and PET / CT in patients with breast cancer.

7. A process for automatic generation of structured medical report for cancer imaging according to claim 1 , 2 or 3 characterized by being configured to an application for CT, MRI and PET / CT imaging in lymphoma cancer where cancer stage follows TNM or Lugano classification system, therapy response evaluation follows Lugano criteria, patient prognosis is predicted using nomograms, A. I. models perform automatic tumor detection and segmentation on PET / CT images, predictions including cancer TNM Stage or Lugano class, total tumor burden, and tumor uptake, among others, clinical data includes cancer history, laboratory tests, previous therapies and it also includes the clinical state of the breast cancer and further technical information about the process of CT or PET / CT image acquisition, where all the information above will make together a radiology report for CT and PET / CT in patients with lymphoma cancer.

8. A process for automatic generation of structured medical report for cancer imaging according to claim 1 , 2 or 3 to fulfil the tasks of data collection of cancer patients, collecting imaging results of cancer patients, and processing collected information to generate a radiological report and create automatically in parallel a database with patient data used to create the report.

Images