Content

Medical informatics is the discipline that deals with the storage, retrieval and optimal use of biomedical information, data and knowledge for problem solving and decision making. Biomedical engineering applies engineering to medicine (devices, imaging, signals), and the two fields overlap. Healthcare 4.0 and digital health: electronic records, telemedicine, wearable monitoring, artificial intelligence, and integration of data across hospitals and patients. Goals: safer and more efficient care, personalised medicine, and better use of resources.

Data are symbols or measurements; information is data interpreted in context. A communication model has a source, encoder (transducer), channel (with noise), decoder and receiver. In medicine, transducers convert physical quantities (temperature, pressure, voltage) into electrical signals. Three aspects of information: syntactic (form and structure of symbols), semantic (meaning) and pragmatic (effect on the receiver's decisions and actions). A good health information system needs all three.

An information-processing system has input (data entry), processing, storage and output. Data entry may be manual (keyboard), automatic (sensors, barcodes) or by speech; errors in entry are a major source of incorrect data. User interfaces: command line, graphical (windows, icons, menus) and touch or voice interfaces. Good interface design reduces errors, which is critical in clinical settings.

1. Define medical informatics. 2. Name the elements of a communication system. 3. Explain syntactic, semantic and pragmatic aspects of information. 4. Why is data entry a critical step?

1. The discipline concerned with storing, retrieving and using biomedical information, data and knowledge for problem solving and decision making. 2. Source, encoder, channel (with noise), decoder, receiver. 3. Syntactic: form; semantic: meaning; pragmatic: effect on decisions and actions. 4. Errors made at entry propagate through every later use of the data.

Stages 1. Acquisition: sensor or electrode, amplifier, analogue filter and analogue-to-digital converter (sampling and quantisation; sampling must respect the Nyquist criterion). 2. Pre-processing: filtering of noise and artefacts (power-line interference, baseline wander, motion). 3. Feature extraction: measurable descriptors such as amplitude, duration, frequency content. 4. Classification or decision: rules or machine-learning algorithms assign the signal to a class (for example normal or abnormal ECG). Examples: ECG, EEG, EMG, blood pressure, pulse oximetry.

Classification by size: PAN (personal), LAN (local), MAN, WAN (wide, for example the Internet). Topologies: bus, star, ring, tree and mesh; the star is the most common in LANs, with a central switch. Media: twisted pair, coaxial cable, optical fibre and wireless (Wi-Fi, Bluetooth). Ethernet (IEEE 802.3) is the dominant wired LAN technology, with frames carrying source and destination addresses. Hospital networks need reliability, security and segmentation of medical devices.

1. List the stages of a biosignal analysis chain. 2. What is aliasing and how is it avoided? 3. Compare bus and star topologies. 4. Give an example of a WAN.

1. Acquisition, pre-processing, feature extraction, classification. 2. Distortion caused by sampling below twice the highest signal frequency; avoided by sampling fast enough and using an anti-aliasing filter. 3. Bus: all devices share one cable and a failure can stop the network. Star: each device connects to a central switch, easier to manage and more robust. 4. The Internet.

A database is an organised collection of related data; a database management system (DBMS) provides definition, storage, querying, security and concurrent access. Three-level architecture: the internal (physical) level, the conceptual (logical) level and the external (user views) level, giving data independence. Entity-relationship (E-R) model: entities (things, for example Patient, Visit, Doctor), attributes (properties) and relationships (associations) with cardinality (one-to-one, one-to-many, many-to-many). Design process: requirements, conceptual E-R design, logical (relational) design, physical implementation.

Relational model: data are stored in relations (tables) of rows (tuples) and columns (attributes). A primary key uniquely identifies each row; a foreign key refers to the primary key of another table and enforces referential integrity. E-R to relational mapping: entities become tables; a many-to-many relationship becomes its own table with two foreign keys. Relational algebra: selection (rows satisfying a condition), projection (chosen columns), union, intersection, difference, Cartesian product and join (combines rows from two tables on a matching condition; natural join uses equal common attributes).

Problem: redundant data cause anomalies on insertion, update and deletion. - First normal form (1NF): all attributes are atomic (no repeating groups). - Second normal form (2NF): 1NF and every non-key attribute depends on the whole key, not part of it. - Third normal form (3NF): 2NF and no non-key attribute depends on another non-key attribute (no transitive dependency). - Boyce-Codd normal form (BCNF): every determinant is a candidate key. Normalisation is achieved by decomposing tables without losing information.

Categories: DDL (CREATE, ALTER, DROP tables), DML (INSERT, UPDATE, DELETE, SELECT) and DCL (GRANT, REVOKE permissions). Basic query: SELECT columns FROM table WHERE condition ORDER BY column. Conditions use comparison and logical operators (AND, OR, NOT), BETWEEN, IN and LIKE. Joins combine tables: SELECT p.name, v.date FROM Patient p JOIN Visit v ON p.id = v.patient_id. Aggregations: COUNT, SUM, AVG, MIN, MAX with GROUP BY and HAVING.

1. What is a foreign key? 2. How is a many-to-many relationship represented in the relational model? 3. Define 3NF. 4. Write a query that lists the patients born after 1990. 5. Difference between projection and selection?

1. An attribute that refers to the primary key of another table and enforces referential integrity. 2. By a separate table with foreign keys to both entities. 3. The table is in 2NF and no non-key attribute depends transitively on the key. 4. SELECT * FROM Patient WHERE birth_year > 1990. 5. Selection chooses rows by a condition; projection chooses columns.

A medical record is the chronological documentation of a patient's health information and care. Organisation: time-oriented (chronological), source-oriented (by department or type of document) and problem-oriented (the SOAP structure: subjective, objective, assessment, plan). The electronic health record (EHR) makes records accessible, searchable and shareable, but raises issues of privacy, security and interoperability.

ICD (International Classification of Diseases, WHO) codes diseases for statistics and reimbursement. ICPC classifies primary care encounters. SNOMED CT is a comprehensive clinical terminology. Lexicons and thesauri map terms to concepts. Other classifications cover procedures and drugs (ATC). Benefits: comparability, research and billing; challenge: choosing the right code and keeping it up to date.

DICOM (Digital Imaging and Communications in Medicine): the standard for medical images, defining both the file format (image plus metadata such as patient and study information) and the network protocol. HL7 (Health Level Seven): messaging standards for clinical and administrative data (HL7 v2 messages, the CDA clinical document architecture). FHIR (Fast Healthcare Interoperability Resources): a modern HL7 standard using web technologies (RESTful APIs) and resources such as Patient, Observation and Encounter.

Components: the hospital information system (HIS), radiology information system (RIS), picture archiving and communication system (PACS), laboratory information system and the electronic record. In Italy, health data flows go to regional and national bodies through the SSN information flows. Goals: continuity of care, reduced errors and support for management and planning.

1. Explain the SOAP structure. 2. What is ICD used for? 3. What does DICOM define? 4. Compare HL7 v2 and FHIR. 5. What is PACS?

1. Subjective, Objective, Assessment, Plan. 2. Coding diseases for statistics, epidemiology and reimbursement. 3. The format and the communication protocol for medical images and their metadata. 4. HL7 v2 uses delimited messages; FHIR uses modern web technologies and resources. 5. Picture archiving and communication system, for storing and distributing medical images.

Big data are characterised by volume, velocity and variety. Machine learning learns patterns from data. Supervised learning uses labelled examples to predict a label: classification (discrete class, for example disease or not) or regression (a number). Common models: logistic regression, decision trees and random forests, support vector machines, neural networks. Evaluation: split data into training, validation and test sets, or use cross-validation. Metrics: accuracy, sensitivity (recall), specificity, precision and the ROC curve. Overfitting means the model memorises the training data and fails on new data; it is controlled by simpler models, more data, regularisation and cross-validation. In medicine, accuracy alone can mislead when a disease is rare.

1. Distinguish classification from regression. 2. What is overfitting? 3. Why is cross-validation used? 4. Why can accuracy be misleading for rare diseases?

1. Classification predicts a category; regression predicts a number. 2. The model fits the training data (including noise) too closely and generalises poorly. 3. To estimate how the model performs on unseen data and to tune it. 4. A model that always predicts no disease has high accuracy but detects no cases.