Doctor’s Fresh Pair of Eyes: How Artificial Intelligence is Transforming Medical Diagnostics

“For instance, AI is successfully applied in the analysis of mammograms, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound and other imaging modalities. It helps detect early signs of cancer, neurological disorders, vascular pathologies and other significant diagnostic indicators,” says Verikas.

However, he emphasises that AI is not designed to replace doctors – it serves as an additional tool that enhances clinical competence and reduces the likelihood of diagnostic errors.

With Every Learning Cycle – More Accurate Diagnosis

In medical image analysis, AI most often relies on deep learning methods, particularly convolutional neural networks (CNNs). These networks are specifically designed for image recognition, making them exceptionally well-suited for complex medical imaging interpretation tasks.

“First, the data must be prepared. Training a model requires large datasets of previously annotated medical images – for example, X-rays with physician-labelled diagnoses or marked pathologies. The AI system then analyses these images, learning to recognise patterns and structures associated with diseases. Each training cycle allows the algorithm to improve its ability to identify diagnostic features accurately,” explains Verikas.

The model is then validated using new, previously unseen images to assess how well it can generate correct diagnoses in real-world conditions. When systematic inaccuracies are identified, the model is further refined and adapted to new data.

“It is important to stress that the quality of AI outcomes directly depends on the quality of the data. Poorly labelled or insufficiently diverse datasets can lead to inaccurate conclusions. Therefore, close collaboration between clinicians, computer scientists and data researchers is essential to ensure reliable and clinically applicable results,” he adds.

Detecting Minute Structural Changes

According to Verikas, AI systems are able to recognise pathologies because of their capacity to analyse vast volumes of imaging data and detect minute structural changes that the human eye might miss or misinterpret.

“CNN models can ‘learn’ to recognise specific microstructural patterns that may be difficult for doctors to identify – such as abnormal cell arrangements, texture variations, tiny tumour markers or calcium deposits in blood vessels. Models like U-Net can accurately segment and localise pathologies, precisely outlining tumour margins or areas of ischaemia in cardiac MRI scans,” he explains.

Moreover, AI can analyse multimodal data – integrating information from different imaging sources, such as CT and MRI scans, to create a more comprehensive and contextual view of a patient’s condition.

“At present, AI algorithms perform particularly well in medical fields where data is abundant and diagnostic standards are clearly defined. Excellent results are achieved, for example, in automating standard cardiac ultrasound measurements, detecting breast cancer in mammograms or identifying pulmonary nodules in CT scans,” notes the SustAInLivWork researcher.

AI is also highly effective in analysing well-visualised structures, such as the retina, for signs of diabetic retinopathy. Reliability remains high in domains with established diagnostic classifications – for instance, evaluating echocardiographic or MRI parameters like ejection fraction or chamber size.

Unlike Humans, AI Never Gets Tired

When it comes to subtle and complex diseases such as early-stage cancers or heart conditions, Verikas highlights several distinct advantages of AI. Firstly, it can process enormous volumes of data that might appear as mere ‘noise’ to the human eye or traditional statistical methods.

“For instance, in cardiology, AI can analyse thousands of CT scans or echocardiograms and detect extremely subtle changes that a doctor might not yet consider pathological, but which could later prove clinically significant. In this way, AI helps to identify diseases at an early stage, when treatment tends to be most effective,” he explains.

Secondly, AI tools offer remarkable consistency. Unlike humans, they do not tire, lose focus or vary in performance due to time of day or mood. This means that an algorithm assessing the same CT or echocardiography image will always produce an identical result. Such stability is particularly important in areas like early detection of oncological changes or the assessment of subtle echocardiographic markers, where doctors’ opinions may sometimes differ.

The Final Diagnosis Always Belongs to the Doctor

However, it is also essential to acknowledge AI’s limitations. According to D. Verikas, AI is not a “magical solution” – its performance depends on the data on which it has been trained. If the training data reflect only a limited patient population, the model may perform less effectively when applied elsewhere. AI can also make mistakes when encountering situations it has not seen before – such as rare anatomical variations, atypical tumours, or lower-quality images. In such cases, the algorithm might provide an incorrect result, even when it is “confident” in its decision.

“Difficulties also arise when diagnostic markers lack clear boundaries or depend on the context of disease progression. Such cases are common in the early stages of Alzheimer’s disease, where an accurate diagnosis requires not only image analysis but also assessment of how the condition evolves over time,” says D. Verikas.

Another limitation is the lack of clinical context. AI can detect early signs of cardiac damage in images but cannot account for a patient’s lifestyle, comorbidities or social factors. It is therefore the physician’s role to integrate AI-derived insights with the broader clinical picture.

“Another challenge is the so-called ‘black box’ problem – it is not always clear why an algorithm makes a particular decision. In medicine, this is critical, as clinicians must be able to understand and explain the reasoning behind a diagnosis. Therefore, the final decision is always made by the doctor,” notes Verikas.

AI Solutions Are Carefully Tested

From an ethical standpoint, the greatest challenge concerns accountability. Patients need to know who made the final decision – the doctor or the algorithm. Another key aspect is data protection.

“Since AI learns from real patient images, it is essential to ensure that this data is properly anonymised and stored in accordance with the General Data Protection Regulation (GDPR) and other legal frameworks. Patients must be confident that their information is used responsibly and will not be misused,” says the researcher involved in the SustAInLivWork project.

AI errors are managed in several ways: algorithms are trained on diverse datasets, continuously tested in real clinical settings, and, most importantly, their results are always reviewed by medical specialists. Experience shows that the best outcomes are achieved when AI acts as a kind of “fresh pair of eyes” for the doctor – helping to spot important details, speeding up analysis, while the doctor ensures clinical context and takes responsibility for the final diagnosis.

“The use of AI in medicine is strictly regulated. Before being introduced into clinical practice, all algorithms must comply with medical device regulations. In Europe, they must receive CE marking, while in the United States they require approval from the Food and Drug Administration (FDA). This means that every system is rigorously tested to ensure it operates safely, reliably and genuinely improves diagnostic quality,” he explains.

According to D. Verikas, AI holds enormous potential to enhance diagnostics, but it must be applied ethically – operating transparently, safely, and always under the supervision of a doctor. Only under such conditions can it become a trusted partner in healthcare rather than a potential source of risk.

SustAInLivWork is the first competence centre of its kind in Lithuania, systematically consolidating knowledge and skills in AI. It brings together four leading Lithuanian universities – Kaunas University of Technology, Vytautas Magnus University, Lithuanian University of Health Sciences, and Vilnius Gediminas Technical University – in partnership with Tampere University (Finland) and Hamburg University of Technology (Germany). It is a long-term, cross-sectoral platform connecting science, business, the public sector, and society.