Eic interpretable medical AI system based on DIKWP artificial consciousness model

Abstract

The invention discloses a medical artificial intelligence decision-making system based on a DIKWP artificial consciousness model. According to the system, a data-information-knowledge-intelligence-intention five-layer cognitive architecture is adopted, and a meta-cognitive feedback mechanism is combined, so that full-link semantic interpretability of a diagnosis and treatment decision process is realized. A medical ethical control unit is arranged in the system to ensure that AI suggestions conform to patient benefits and industry specifications; and a semantic BUG detection module is arranged to monitor the reasoning deviation so as to improve the safety. Through a natural language interaction interface, doctors and patients can understand AI decision basis and provide feedback, and patient preferences are taken into consideration of decision. Besides, the system is integrated with the traditional Chinese medicine overall diagnosis and treatment concept, realizes integration of traditional Chinese medicine knowledge and western medicine knowledge while providing personalized, explainable and safe clinical decision support, and can be used for scenes such as clinical diagnosis and treatment decision support and the like.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and medical information technology, and in particular to a medical decision support system embedding an artificial consciousness model. This system possesses interpretability and ethical constraints, and can be used in scenarios such as hospital clinical decision support and intelligent diagnosis and treatment assistance, ensuring that AI-generated diagnostic and treatment recommendations are semantically transparent, safe, and ethically compliant. Background Technology

[0002] Currently, AI technologies such as Large Language Models (LLMs) have shown great potential in medical diagnosis and decision-making, but their "black box" decision-making process and potential ethical risks have raised concerns. Traditional AI systems often fail to make their decision-making logic understandable to humans, leaving patients and doctors lacking the right to know and making it difficult to trust the treatment recommendations provided by AI. Furthermore, AI may exhibit behaviors inconsistent with human values ​​(such as recommending solutions that are unethical or contrary to patient interests). Existing technologies lack a mechanism to make the AI ​​decision-making process completely transparent and traceable, and to embed medical ethics and value constraints internally to prevent adverse decisions. Therefore, there is an urgent need for a medical AI system that can make the AI ​​decision-making process transparent and explainable and incorporates ethical constraints to fundamentally solve the above problems. Summary of the Invention

[0003] This invention provides a medical AI system architecture based on the DIKWP (Digital Intelligence, Knowledge, Wisdom, Intent) artificial consciousness model. By introducing a five-layer network cognitive structure of Data-Information-Knowledge-Wisdom-Intent (DIKWP) and a metacognitive loop, it achieves end-to-end semantic interpretability and self-regulation of AI decision-making. Specifically:

[0004] DIKWP Mesh Semantic Architecture divides the AI ​​reasoning process into five semantic stages: data acquisition, information processing, knowledge association, intelligent decision-making, and intent guidance, connecting each layer through a mesh-like two-way feedback mechanism. Each step has clear semantic definitions and indicators, ensuring that every AI decision-making process is traceable, explainable, and understandable to humans. For example, for each diagnostic conclusion, the system can provide the data and knowledge evidence on which it is based, as well as the reasoning path that conforms to medical principles. Doctors and patients can view the visualized representation of the AI's internal knowledge graph and reasoning links, achieving white-box visualization analysis, rather than just seeing the final conclusion.

[0005] A Conscious Unit of Ethical and Value Constraints: The DIKWP architecture introduces an "Intent / Purpose" layer as a built-in ethical control unit. This layer pre-integrates medical ethical principles and patient value preferences, ensuring that AI decisions always serve the patient's interests and safety needs. For example, the system has built-in informed consent rules; when AI provides a treatment plan, it checks whether the patient understands the relevant risks and benefits; if there is a tendency to deviate from human values, adjustments are immediately made to ensure that the decision is ethical.

[0006] "Consciousness BUG" Early Warning Mechanism: This invention innovatively sets up a semantic vulnerability (BUG) detection module for AI self-monitoring. This module continuously monitors the semantic state of the five-layer cognitive link. If abnormal cognitive biases or unreasonable inferences (similar to human subjective assumptions, hallucinations, and other cognitive defects) are detected, an early warning and correction mechanism is triggered. For example, when the AI's reasoning exhibits hallucinations (inferences not based on facts) or intentional deviations (deviating from the expected medical goals), the semantic firewall will immediately issue a warning and block the propagation of the error. This mechanism simulates the self-correcting ability of human consciousness, allowing for a certain degree of "imperfection" but preventing serious errors.

[0007] Semantic feedback and patient informed consent interaction: The system provides a human-computer interaction interface to explain the rationale behind decisions to doctors and patients in natural language and obtain feedback. Patients can ask about the basis of the AI ​​diagnosis, and the AI ​​will extract key information from the five-layer semantic structure and provide a simple explanation, ensuring the patient's right to know. If the patient expresses a personal preference (e.g., a preference for conservative treatment), the AI ​​will treat this as a new intention input and adjust its decisions to respect the patient's values. The entire process forms a closed loop of two-way semantic communication between AI and humans, improving the acceptance and personalization of treatment plans.

[0008] Integration of Traditional Chinese Medicine (TCM) Cultural Elements: TCM's holistic perspective and other cultural elements are incorporated into the ethical framework and semantic model. For example, system decision-making not only considers biological indicators but also references TCM concepts such as "body-mind unity" and "harmony between man and nature" to conduct a holistic assessment of the patient's condition. This integration makes AI decision-making more aligned with Eastern medical ethics (such as benevolence and skillful practice, and the concept of "treating disease before it occurs"), emphasizing prevention and aligning with proactive medicine principles. Simultaneously, it draws on the transparent diagnostic thinking mode of TCM's "observation, auscultation, inquiry, and palpation" to enhance the intuitiveness and user-friendliness of AI explanations.

[0009] Through the above design, the medical AI system constructed in this invention achieves highly interpretable and controllable diagnostic and treatment decision support. Each step of reasoning is verifiable, meeting regulatory requirements for AI safety while increasing trust in AI among doctors and patients. This system architecture is expected to become the foundational hub for next-generation medical AI, providing an innovative solution to the black-box problem of large-scale medical applications. Furthermore, the system can interface with international medical AI ethical standards and regulations (such as the WHO guidelines on artificial intelligence ethics) as needed, supporting adjustments to strategy modules to meet the regulatory requirements of different countries. Overall, this invention combines technological advancement with clinical practicality, demonstrating promising clinical application prospects.

[0010] Compared with the prior art, the medical AI system of the present invention has the following beneficial effects:

[0011] Semantic interpretability: Through the DIKWP five-layer semantic architecture and knowledge graph visualization, every step of the reasoning process is traceable, and AI can explain "why and how" a certain diagnosis / decision is reached. Doctors and patients can view key data evidence, related medical knowledge, and the complete reasoning chain, completely breaking away from the traditional AI "black box" mode and significantly improving the transparency and understandability of decision-making.

[0012] Built-in ethical constraints: This system directly embeds medical ethical principles and patient values ​​into the AI ​​decision-making process (intent layer), ensuring that the output meets clinical ethical requirements and patient interests. Ethical review and adjustment are conducted within the decision-making process itself, which is more proactive and effective than post-hoc manual oversight. This design makes AI recommendations more trustworthy and adaptable to the requirements of medical regulations or guidelines in various countries, achieving a balance between compliance and humanistic care.

[0013] Safety and Robustness: The innovative semantic bug detection module adds "double insurance" to AI decision-making. This module continuously monitors the reasoning chain, and immediately alerts and triggers corrections upon the occurrence of cognitive errors, hallucinations, or abnormal deviations, preventing errors from accumulating and propagating to the final decision. Even if the AI ​​model itself has a certain degree of uncertainty, the system can still ensure the safety and reliability of the final output solution by relying on metacognitive loops and self-correction mechanisms, significantly reducing the risks of AI diagnosis and treatment.

[0014] Patient-centered interaction: The system's human-computer interface supports natural language interpretation and incorporates patient feedback, enabling two-way communication between medical AI and patients. Patients can fully understand the basis of AI suggestions, helping to eliminate doubts and participate in decision-making; at the same time, patients' personal preferences and opinions are promptly adopted by AI to adjust treatment plans, improving the personalization of treatment plans and patient satisfaction. Doctors can also supervise and guide AI through interaction, enhancing their sense of control over AI assistance and promoting acceptance and trust between doctors and patients.

[0015] Cultural Compatibility: This invention incorporates the holistic approach and knowledge of Traditional Chinese Medicine (TCM), making AI decision-making more aligned with medical practices within an Eastern cultural context. Through the integration of TCM and Western medical knowledge, the system can provide diverse treatment recommendations to meet the needs of different patients. In domestic deployment, the inclusion of TCM elements helps increase patient acceptance of AI recommendations; in international applications, the system architecture also possesses the ability to expand and be compatible with other traditional medical systems or local ethical requirements, demonstrating broad adaptability.

[0016] In summary, this invention effectively addresses the pain points of unexplainable decision-making, ethical deficiencies, and poor cultural adaptability in the field of medical AI. By combining AI's powerful reasoning capabilities with the humanistic spirit of medical care, it significantly enhances the practical value and reliability of AI in clinical decision support. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the DIKWP five-layer cognitive process in the system of the present invention.

[0018] Figure 2 This is a schematic diagram of the semantic feedback mechanism and ethical control architecture of the system of the present invention. Detailed Implementation

[0019] like Figure 2 As shown, the medical AI system of this invention includes five cognitive processing units and several auxiliary modules. The five cognitive processing units are, in order, a data processing layer, an information processing layer, a knowledge association layer, an intelligent decision-making layer, and an intent layer, which together constitute the core of the DIKWP artificial consciousness model. The functions of each layer are progressively advanced: the data processing layer is responsible for collecting and preprocessing patient-related raw data (such as symptom descriptions, physical signs measurements, examination and test results, etc.); the information processing layer transforms the raw data into structured information (e.g., extracting keywords and key points of medical history from textual symptoms); the knowledge association layer uses a medical knowledge base to associate the extracted information with existing medical knowledge (e.g., retrieving possible disease diagnoses and treatments based on symptoms and test results); the intelligent decision-making layer makes a comprehensive judgment based on the knowledge provided by the previous layer, forming a preliminary diagnosis and treatment decision plan; and the intent layer guides the final decision output and reviews and adjusts the plan in conjunction with ethical controls.

[0020] In addition to the five core units mentioned above, the system also includes the following modules: an ethics control unit that works in conjunction with the intent layer to apply medical ethics rules and patient preference constraints during the decision-making process; a semantic bug detection module to monitor the semantic state of the five-layer inference chain in real time, detect abnormal deviations, and trigger warnings and corrections; a user interaction interface to provide a natural language interaction channel between doctors / patients and the AI ​​system, enabling bidirectional semantic communication of explanation output and feedback input; and a medical knowledge base to store massive amounts of medical knowledge supporting diagnostic decisions. The medical knowledge base can take the form of a knowledge graph, containing knowledge of diseases, symptoms, drugs, and guidelines from the Western medicine field, as well as knowledge of syndromes, prescriptions, and treatment experience from the traditional Chinese medicine field, achieving the integration and association of knowledge from both Western and traditional Chinese medicine, and providing diversified information support for the knowledge association layer and the intelligent decision-making layer.

[0021] In this system, the AI's decision-making reasoning unfolds sequentially according to the five stages of DIKWP, such as... Figure 1As shown, the process is as follows: First, the data layer acquires the patient's raw data, including consultation dialogue text, vital signs, and medical images, and performs necessary cleaning and encoding to form machine-processable input. Then, the information layer analyzes and extracts data, such as identifying names, medical terms, and symptom indicators, transforming the chaotic data into a meaningful set of information. Next, the knowledge layer retrieves relevant medical knowledge and evidence from a knowledge base, such as matching possible diseases based on patient symptoms, referencing treatment plans in clinical guidelines, or retrieving historical cases and traditional Chinese medicine diagnostic knowledge. Subsequently, the intelligence layer synthesizes multiple candidate conclusions and solutions from the knowledge layer, performing reasoning and decision optimization, such as comparing the probability of different diagnoses and the risk-benefit of different treatment options, to form the optimal decision recommendation. Finally, the intent layer receives the preliminary decision results from the intelligence layer and, in conjunction with preset medical goals and ethical guidelines, guides and modifies the plan, outputting the final treatment recommendation.

[0022] The above five stages are not a simple one-way pipeline; information flows bidirectionally between layers through a semantic feedback mechanism. For example, when a higher layer finds that the information provided by a lower layer is insufficient or contradictory, it can request supplementary data or adjustments to the analysis methods. Similarly, the intent layer checks whether the solution provided by the wisdom layer aligns with the patient's value preferences, and if necessary, requests the wisdom layer to reconsider the solution. This metacognitive loop allows the AI ​​system to iterate multiple times during a single reasoning process: if the initial decision obtained by the system fails to pass the goal and ethical review of the intent layer, it will feed back adjustment signals to the relevant processing units at the lower layers (such as the wisdom layer and knowledge layer) for recalculation; this process repeats until a decision that meets the conditions is generated. Formally, the reasoning process at each layer can be abstracted as a combination of functions: Let D, I, K, W, and P represent the internal semantic states of the data layer, information layer, knowledge layer, wisdom layer, and intent layer, respectively, then the reasoning flow can be represented as:

[0023] i=f1(D),K=f2(I),W=f3(K),P=f4(W)

[0024] Where f1, f2, f3, and f4 are the inference transformation functions for each layer. Under the influence of metacognitive feedback, the above process is executed iteratively: for example, a feedback correction function f5 is defined to adjust the state W of the intelligence layer according to the current intention output P, ​​denoted as W. ′ =f5(P,W), then recalculate the new intent output P. ′ =f4(W ′ Through such feedback updates, the system gradually approaches the optimal solution. In the implementation, it can be set to stop iterating and output the result when the decision scheme changes little in two consecutive iterations or satisfies all constraints.

[0025] This system also establishes the concept of inference path weights to quantify the reliability of each inference chain. To this end, a credibility weight ω can be introduced for each stage of the inference process. i (i = 1, ..., 4), corresponding to the four transformation stages from the data layer to the intent layer. The overall weight of the entire decision chain can be defined as the product of the weights of each stage:

[0026]

[0027] A higher weight Ω indicates that the reasoning path is more supported and the result more credible. If the overall weight of a path is too low, it means that there is significant uncertainty or weak support in a certain step of the reasoning. This path will be marked as unreliable and will not be adopted or will require further review. On the interpretation interface, the system can also present the weight distribution to doctors to indicate which evidence contributes most to the conclusion and which steps have uncertainty.

[0028] The Ethics Control Unit is a crucial module ensuring the system's "human-centered" approach and adherence to safety bottom lines. Embedded at the top layer (intent layer) of the DIKWP architecture, this unit conducts ethical reviews and guides the purpose of AI-generated decisions. Internally, it contains a professional medical ethics rule base and value assessment model. For example, the rule base includes constraints related to the four principles of medical ethics (respect for patient autonomy, beneficence, benefit, and fairness): the informed consent rule ensures that any treatment plan is fully considered in light of the patient's understanding and consent; the patient benefit maximization rule ensures that the best option for the patient's overall benefit is prioritized; the non-maleficence principle avoids measures that may cause unacceptable harm to the patient; and the fairness principle ensures that patients are not discriminated against due to non-medical factors. If the AI's initial decision violates any of these principles, the Ethics Control Unit will intervene promptly to correct or restrict it.

[0029] Furthermore, the ethics control unit can adjust decisions based on individual patient value preferences. The system extracts specific requirements or preferences from patient interactions (e.g., a patient explicitly rejects a certain invasive procedure and prefers conservative treatment first) and stores these preference parameters in the ethics control unit. During the decision-making process, the intent layer considers patient preferences as part of the decision objective: that is, the AI ​​not only seeks the medically optimal solution but also strives to meet the patient's wishes as much as possible. In practice, this can be achieved by adjusting the AI's objective function—for example, by incorporating an evaluation of patient satisfaction / preference into the decision scoring. If the AI's proposed solution is inconsistent with the patient's pre-declared wishes, the ethics control unit will lower the solution's score or directly request the generation of an alternative. Through this mechanism, the system ensures that the final output recommendation is not only medically reasonable but also acceptable and trustworthy to the patient.

[0030] The "consciousness bug" detection module in AI systems simulates the self-monitoring and error-awareness capabilities of the human brain, aiming to promptly detect semantic loopholes or anomalies in the AI ​​reasoning process. These anomalies include, but are not limited to: factual illusion (making unfounded guesses about non-existent conclusions), logical jumps (illogical breaks in the reasoning process), cognitive biases (inappropriate tendencies due to biased training data), and intentional deviations (system behavior deviating from established medical goals or ethical guidelines).

[0031] This module embeds monitoring points at each layer of AI inference to continuously examine the semantic content of the output from each layer. When an anomaly is detected, the module issues an alarm signal and triggers corresponding corrective processing. For example, if the intelligent decision-making layer gives a conclusion lacking knowledge support (suspected hallucination), the bug detection module will issue an alert within milliseconds, suspend the current decision output, and report the problem to the knowledge layer to request supplementary evidence or to the information layer to retrieve key evidence again. Similarly, if the intent layer finds that the current decision is seriously inconsistent with the expected medical goal (such as suggesting a treatment plan with obvious potential harm to the patient), the bug detection module will treat it as intent drift, immediately block the plan, and notify the ethics control unit to intervene.

[0032] To achieve the aforementioned monitoring, this module employs techniques such as semantic similarity calculation and abnormal pattern recognition. Specifically, the output of each cognitive layer can be represented as a vector form v. D ,v I ,v K ,v W ,v P These correspond to the semantic representations from the data layer to the intent layer, respectively. Then, the semantic consistency between the outputs of adjacent layers is calculated, for example, using cosine similarity.

[0033]

[0034] If the similarity C between a pair of adjacent layers (such as the knowledge layer and the wisdom layer) is lower than a preset threshold, it indicates that the reasoning from that layer to the next layer has introduced too much new information or deviated from the original semantic range, i.e., a suspicious reasoning jump has occurred. At this time, the module determines that there is a potential cognitive bug and generates an alarm. In addition to semantic similarity, the module can also combine other indicators (such as comparison with the medical common sense base, built-in logical rule verification, etc.) for comprehensive judgment. While issuing an alarm, the system will take different corrective strategies according to the type of bug: for general reasoning inconsistencies, a metacognitive loop is triggered to adjust the lower-level reasoning through feedback from higher levels; for serious errors (such as conclusions that obviously violate medical common sense), the output is directly blocked and a manual review is required. Through multi-level monitoring and intervention, the "awareness bug" module significantly improves the robustness of AI decision-making, ensuring that even if a certain degree of bias occurs within the AI, it will not evolve into an error that endangers patient safety.

[0035] The user interaction interface module of this invention is responsible for connecting the AI ​​system with doctors / patients, building a transparent communication bridge between them. On the one hand, the interface accepts natural language input from doctors or patients, including asking questions, requesting explanations, providing personal preferences, and even new symptom information; on the other hand, it feeds back the decision-making basis and process of the AI ​​system to the user in a human-understandable form, achieving semantic-level communication.

[0036] The interactive interface can integrate Natural Language Processing (NLP) technology to understand user questions or requests. For example, when a patient asks, "Why is this treatment recommended?", the interface converts the question into an internal semantic query, guiding the AI ​​system to extract the reasons behind the decision. The AI ​​system extracts key explanatory elements from the DIKWP five-layer structure, including relevant data evidence, medical knowledge, and reasoning chains, and then responds to the patient in concise and clear language through the interface. This explanation process is equivalent to the AI ​​"telling a story to the patient," explaining how it starts from the symptoms, what knowledge it uses, and what logic it follows to arrive at the current recommendation. For doctors or professionals, the interface can also provide more technically detailed explanations, such as visualizing the reasoning path using a knowledge graph: nodes represent knowledge units such as symptoms, test results, disease diagnoses, and treatment plans, and lines represent reasoning connections, allowing doctors to intuitively review the AI's thought process.

[0037] The interactive interface also supports integrating user feedback into AI decision-making. After understanding the AI's suggestions and rationale, patients or doctors can express further opinions or preferences through the interface. For example, a patient might ask, "I'd prefer to try medication first and don't want surgery immediately, is that okay?" The interface will translate this preference into semantic modification instructions within the AI ​​and transmit them to the intent layer. Upon receiving the new "intent" input, the intent layer will adjust the direction of the AI's decision (e.g., favoring conservative treatment), and then use a metacognitive loop to re-plan the treatment plan at the lower levels. The entire process forms a closed loop of human-machine collaboration: the AI ​​not only explains its own decisions but also listens to human thoughts in real time and adjusts accordingly. This interactive mode ensures the patient's right to know and participate, making the patient part of the decision-making process and increasing the acceptability of the AI's proposed solutions. For doctors, this two-way semantic interaction also provides a greater sense of control and trust—doctors can question the AI ​​at any time about "why it did this" and supervise and guide the AI's thinking.

[0038] A key feature of this invention is the integration of Traditional Chinese Medicine (TCM) knowledge and Eastern medical concepts into an artificial intelligence decision-making framework, thereby achieving intelligent diagnosis and treatment that combines TCM and Western medicine. This integration is reflected in several aspects:

[0039] First, at the knowledge base level, the system's medical knowledge base simultaneously incorporates both Western and Traditional Chinese Medicine (TCM) knowledge systems. Western medicine knowledge includes evidence-based diagnostic criteria, treatment guidelines, and drug mechanisms of action; TCM knowledge covers TCM theories (such as the theory of Zang-Xiang and meridians), knowledge of Chinese herbal medicines and prescriptions, historical medical records, and experience. During reasoning, the knowledge association layer simultaneously queries both knowledge bases. For example, when a patient presents with a certain set of symptoms, the Western medicine knowledge base might return several differential diagnoses of Western diseases, while the TCM knowledge base would return the corresponding TCM syndrome (such as "wind-cold cough" or "phlegm-dampness accumulation in the lungs") and recommended prescriptions. At the intelligent decision-making layer, the system comprehensively considers this information: analyzing both the physiological and pathological factors of the Western medicine diagnosis and assessing the overall state of the TCM syndrome differentiation, thus providing a more comprehensive understanding.

[0040] Secondly, the system incorporates the holistic and preventative concepts of Traditional Chinese Medicine (TCM) into its decision-making and reasoning approach. TCM emphasizes the unity of man and nature and the holistic approach to mind and body. Patient assessments often focus not only on local symptoms but also on overall health, living environment, and emotional factors. This system incorporates these dimensions at the information processing layer, such as obtaining patient information on sleep, emotions, and diet through questionnaires or sensors, and using this information as part of the decision-making process. This allows the AI ​​to simulate the TCM physician's diagnostic methods of observation, auscultation, inquiry, and palpation when making decisions at the intelligent level, taking a holistic approach and avoiding treating symptoms rather than the root cause. The TCM concept of "prevention is better than cure" is also reflected in the treatment plan—when providing a treatment plan, the AI ​​not only focuses on curing the current disease but also incorporates preventative healthcare suggestions. For example, for patients with recurrent chronic diseases, the system may add TCM healthcare options (such as dietary therapy, exercise, and acupuncture suggestions) in addition to standard treatment to improve the patient's long-term physical condition and immunity, reducing disease recurrence. This proactive prevention and long-term conditioning approach aligns perfectly with the modern medical concept of chronic disease management, providing more humane care.

[0041] Furthermore, in terms of explanation and presentation, the unique language and analogies of Traditional Chinese Medicine (TCM) are used to enhance the accessibility of AI explanations. For patients, the system sometimes uses TCM concepts to help understand complex medical problems. For example, it might use "internal balance imbalance" to explain certain pathological mechanisms, or "seasonal regulation" to illustrate health preservation advice, making it easier for patients to accept. Simultaneously, for professionals, the system can output the TCM diagnostic and treatment basis, such as indicating the patient's current syndrome type, typical symptoms, and why a particular prescription is used according to TCM theory. This explanatory information allows TCM practitioners to accept the AI's decisions, making the AI ​​suggestions truly acceptable to medical personnel from different backgrounds.

[0042] Through this multi-layered integration, this system can provide more comprehensive and culturally compatible medical AI services. When dealing with patients who prefer Traditional Chinese Medicine (TCM), the AI's proposed solutions can incorporate TCM therapies, increasing patient trust and satisfaction. In Western-dominated clinical environments, introducing TCM elements can also provide additional insights, particularly in areas such as chronic disease and sub-health management. This intelligent system combining TCM and Western medicine not only expands the breadth of AI knowledge but also reflects respect and inclusiveness towards different medical cultures.

[0043] The following section uses a specific medical scenario to simulate the usage of the system of this invention, including the interaction of input and output and the details of internal system processing.

[0044] Scenario: Ms. Wang, 45 years old, has recently been experiencing recurrent sore throat and cough. She has taken cold medicine on her own, but her symptoms have not improved significantly, and she also experiences fatigue and poor appetite. Ms. Wang prefers to use traditional Chinese medicine and is reluctant to use antibiotics initially. The AI-assisted diagnosis and treatment system of this invention is now assisting the doctor in developing a treatment plan for her.

[0045] Medical Information Acquisition: Ms. Wang described her symptoms through the human-computer interaction interface: "I've had a sore throat and a slight cough for the past two weeks, and I feel very tired." The system's data processing layer also retrieved her medical history and most recent blood test results (showing mild inflammation) from the hospital's electronic medical records. Simultaneously, the patient expressed a preference for first-line treatment with traditional Chinese medicine and a reluctance to immediately take antibiotics. This information, after preprocessing by the data layer, forms a structured input dataset.

[0046] Information Processing: The information processing layer analyzes the acquired text and numerical data. It extracts symptom keywords such as "sore throat," "cough," "fatigue," and "poor appetite," as well as time characteristics such as "lasting for two weeks" and important examination indicators such as "slightly elevated white blood cell count." It also identifies patient preferences such as "preferring traditional Chinese medicine and having a low willingness to use antibiotics." This extracted information is organized into standardized information objects, such as symptom lists, examination result lists, and patient preference parameters, for downstream use.

[0047] Knowledge Retrieval: The knowledge association layer retrieves relevant knowledge from the medical knowledge base based on the above information. First, in the Western medicine knowledge sub-base, possible diseases are retrieved, such as "chronic pharyngitis" and "bronchitis," along with corresponding standard treatment plans (gargling for chronic pharyngitis, antibiotic therapy for bronchitis, etc.). Simultaneously, in the Traditional Chinese Medicine (TCM) knowledge sub-base, possible TCM syndromes are retrieved based on symptom combinations, such as "lung meridian wind-heat" (corresponding symptoms: sore throat, cough) and "qi deficiency constitution" (corresponding symptoms: fatigue, poor appetite). The TCM database also provides relevant prescription recommendations, such as Yinqiao Powder for lung meridian wind-heat, combined with Sijunzi Decoction for regulating qi deficiency. The knowledge layer summarizes the retrieval results, including the basis, typical manifestations, and treatment measures for each candidate diagnosis / syndrome, forming a set of knowledge entries that can be used by the decision-making layer. A preliminary knowledge graph is also constructed: connecting patient symptoms with the aforementioned disease / syndrome nodes, and diseases with potential treatment plan nodes, providing a reference for subsequent reasoning.

[0048] Intelligent Decision-Making: The intelligent decision-making layer comprehensively evaluates multiple possible diagnostic and treatment options from the knowledge layer. The system first reasons about Western medicine diagnoses: for example, based on Ms. Wang's symptom duration and inflammatory markers, it leans towards a diagnosis of "chronic pharyngitis," inferring that the cause might be bacterial infection, with a standard treatment of 7 days of oral broad-spectrum antibiotics. However, the system also assesses that the patient's symptoms are not very severe and she can tolerate conservative treatment; combined with the patient's preference for not immediately using antibiotics, the system considers the direct use of antibiotics to have lower scores in terms of patient acceptance and necessity. Next, the system evaluates Traditional Chinese Medicine (TCM) options: TCM diagnosis indicates Ms. Wang has "wind-heat invading the lungs, with qi deficiency and incomplete recovery," recommending first using heat-clearing and detoxifying Chinese herbal medicines to treat the sore throat, followed by qi-tonifying formulas to regulate the body's constitution. The advantage of this approach is that it aligns with the patient's preferences and has fewer side effects; the disadvantage is that it may be slightly slower to take effect. The intelligent layer scores each option based on a built-in decision-scoring model, considering indicators such as efficacy, safety, and compliance. The results showed that the traditional Chinese medicine conservative treatment plan scored high in terms of patient compliance and safety, while the antibiotic plan was slightly better in terms of speed of efficacy, but the overall score was slightly lower (because patient resistance reduces the actual efficacy). Therefore, the wisdom layer tends to choose the treatment plan based on traditional Chinese medicine, while retaining the Western medicine plan as an alternative.

[0049] Ethical Intent Verification: The intent layer conducts ethical and goal-oriented reviews of the solutions selected by the wisdom layer. First, it checks whether the solution maximizes patient benefit: Although the traditional Chinese medicine solution is slightly slower to take effect, it is expected to cure the patient with minimal side effects, thus aligning with the patient's interests and avoiding unnecessary antibiotic use, adhering to the "non-harm" principle; patient preferences are also respected, adhering to the "respect for autonomy" principle. The ethics control unit determines the solution is ethically acceptable. Next, the intent layer uses "patient recovery and satisfaction" as the ultimate intent goal and assesses whether the current solution needs adjustment: Considering that antibiotics, as an alternative, might be faster in some situations, and that restarting the solution if the patient's condition worsens later would pose a risk, the intent layer decides to add a contingency plan to the solution—instructing the patient to return for a follow-up visit to consider antibiotic use if symptoms do not improve within three days. This adjustment ensures that the solution respects the patient's current wishes while also providing countermeasures for potential risks. After the above verification and adjustments, the final decision-making solution is finalized at the intent layer.

[0050] Output and Interpretation: Through the user interface, the AI ​​provides feedback on the final treatment plan to the responsible doctor and patient. For the doctor, the system displays a visual knowledge graph and text description on the interface: the graph marks symptom nodes, possible diseases and syndrome nodes, as well as the selected treatment plan node; the text description details the recommended plan: "Traditional Chinese Medicine Treatment + Observation," specifically "Administer Yinqiao Powder combined with Sijunzi Decoction, modified, for 7 days of traditional Chinese medicine treatment, while strengthening rest and nutrition; antibiotics are not used for the time being. If there is no significant improvement in symptoms after 3 days, antibiotic treatment will be considered." The explanation is: "This plan is recommended because the patient's infection is relatively mild and the patient prefers traditional Chinese medicine treatment. The traditional Chinese medicine plan can effectively relieve symptoms and enhance the body's resistance, improving the patient's comfort; at the same time..." To avoid the potential side effects and drug resistance risks of antibiotic overuse, we have prepared an antibiotic regimen in case the treatment is ineffective in the short term to ensure therapeutic efficacy. For Ms. Wang, the system pushed a simple explanation via the mobile app: "Your sore throat is mainly caused by inflammation. Traditional Chinese medicine can help clear heat and detoxify, and boost immunity. It may take a few days to see results. Please drink plenty of water and rest during this time. We are not using antibiotics for now because your condition is not serious; a gentle approach is the safest. If it doesn't improve after three days, we will consider other methods." After reviewing the explanation, the patient expressed understanding and acceptance, and felt relatively confident in the traditional Chinese medicine treatment plan.

[0051] Patient Feedback and Adjustments: During the treatment, Ms. Wang could provide feedback on her feelings at any time through the interactive interface. For example, two days later, she reported that "her cough has significantly lessened, and her throat is not so sore anymore." The system's data layer receives this new data, the information layer updates the symptom status, and the knowledge layer assesses that the condition has improved and no change in treatment plan is needed. Based on this, the intelligence layer decides to maintain the original plan, and the intent layer reconfirms that it meets the expected goals. As the patient continues to provide feedback, the AI ​​system continuously updates its confidence level in the condition and notifies the doctor of these changes through the interface, allowing the doctor to understand the treatment progress and feel at ease with AI-assisted monitoring. Ultimately, with human-machine collaboration, Ms. Wang recovered. The entire process demonstrates how the system of this invention operates in actual clinical practice: from data collection, knowledge reasoning, ethical review to interactive interpretation and feedback, every step is transparent and traceable, and fully considers the patient's wishes and cultural background.

Claims

1. An ethically explainable medical AI system based on the DIKWP artificial consciousness model, characterized in that: The system comprises five sequentially connected cognitive processing layers (data processing layer, information processing layer, knowledge association layer, intelligent decision-making layer, and intent layer), an ethics control unit connected to the intent layer, a semantic bug detection module for monitoring the reasoning process of the above five layers, a user interaction interface, and a medical knowledge base. The five cognitive processing layers are connected to form a network structure through a bidirectional semantic feedback mechanism, realizing a metacognitive loop of step-by-step reasoning from data to intent and self-correction. The ethics control unit has built-in medical ethics rules and patient preference information, which are used to impose ethical constraints and guide the purpose of AI decision-making at the intent layer, ensuring that the output results comply with medical ethics principles and patient interests. The semantic bug detection module is used to monitor the semantic consistency in the five cognitive links in real time. When abnormal cognitive deviations are detected, an alert is triggered and error propagation is blocked to perform decision correction. The user interaction interface is used to receive natural language queries and preference inputs from doctors or patients, and convert them into semantic information to provide to the intent layer. At the same time, it feeds back the key evidence and reasons in the AI ​​decision-making process to the user in a human-understandable form. The medical knowledge base contains knowledge data in the fields of Western medicine and traditional Chinese medicine, providing integrated knowledge support for the knowledge association layer and the intelligent decision-making layer.

2. The AI ​​system according to claim 1, characterized in that: The ethics control unit has a built-in medical ethics rule base, including constraints based on principles such as informed consent, the best interests of the patient, no harm, and fairness. When the decision-making scheme generated by the intent layer does not conform to the above ethical rules, the ethics control unit automatically modifies or restricts the decision. The ethics control unit also records and updates the patient's personal preference parameters, and adjusts the AI's decision objective function accordingly during the decision-making process so that the final solution meets the patient's autonomous wishes and special needs.

3. The AI ​​system according to claim 1, characterized in that: The semantic bug detection module monitors the inference link by calculating the semantic similarity or consistency index between the output content of each cognitive processing layer. When the semantic similarity between adjacent layers is lower than a preset threshold, it determines that there is a cognitive bias, thereby triggering an alarm signal and starting the inference correction process of the relevant layer.

4. The AI ​​system according to claim 1, characterized in that: The user interaction interface includes a natural language processing unit for converting doctor's or patient's inquiries into semantic input that AI can understand, and an explanation generation unit for converting the reasoning links and knowledge bases within the AI ​​system into human-understandable language or visual information and outputting it to the user. The user interaction interface allows patients to provide feedback preferences in real time and send this information as new semantic input to the intent layer, thereby dynamically influencing the AI's decision-making process and forming a closed-loop interaction involving the patient.

5. The AI ​​system according to claim 1, characterized in that: The medical knowledge base is implemented by a structured medical knowledge graph, which includes two parts: Western medicine clinical knowledge and traditional Chinese medicine knowledge, and is associated under a unified semantic model. During the reasoning process, the knowledge association layer simultaneously queries the Western medicine disease diagnosis knowledge nodes and traditional Chinese medicine syndrome and prescription knowledge nodes in the knowledge graph, and provides decision support by integrating the diagnosis and treatment suggestions of these two medical systems based on the patient's specific information.

6. The AI ​​system according to claim 1, characterized in that: The system iteratively corrects the lower-level reasoning results through the intent layer to achieve metacognitive cyclical decision-making. After obtaining the initial decision scheme, the intent layer evaluates its deviation from the expected medical goals and ethical principles. If a deviation exists, it generates a feedback adjustment signal and sends it to the intelligent decision layer and / or knowledge association layer for re-reasoning. This iterative process continues until the decision scheme meets the predetermined consistency requirements and ethical constraints.

Images