Page 52 - AIH-2-3

Artificial Intelligence in Health AI in medical diagnostics: A multi-disease approach

of receiver operating characteristic (ROC) curves. diagnostic accuracy across demographic groups, including
Classification accuracy measured the proportion of correct age, gender, and ethnicity. Bias mitigation techniques, such
predictions, while sensitivity quantified the ability to detect as fairness-aware algorithms and regularization methods,
true positive cases. were applied to reduce disparities. Ethical considerations
Specificity assessed the accuracy in identifying true included patient data privacy, the implications of AI-driven
negative cases, and precision captured the proportion diagnostic errors, and the need for regulatory frameworks
that support AI integration into healthcare while protecting
of true positives relative to all positive predictions. The patient welfare. These measures ensure the responsible
F1-score offered a very balanced metric that combines deployment of AI within the sensitive domain of medical
both precision and recall, making it particularly useful diagnostics, emphasizing the importance of accuracy,
for evaluating performance on the various types of equity, and transparency.
imbalanced datasets. The AUC-ROC provided an overall
assessment of model performance across different 3. Background research on available
classification thresholds. These results were benchmarked knowledge
against traditional diagnostic methods to gauge potential
improvements in diagnostic accuracy, efficiency, and speed AI in healthcare represents a significant technological
across the four disease categories. advancement, simulating human cognition to analyze,
interpret, and present complex medical and healthcare
The experiments were conducted using high- data. This ability not only mimics human thought processes
performance computing infrastructure capable of but also enhances healthcare delivery by enabling faster
managing extensive medical datasets and complex model and more accurate diagnoses, treatments, and preventive
architectures. The process involved three main phases. measures. ML and DL algorithms, key components of
First, data preprocessing included image normalization, AI, can process vast amounts of clinical data, such as
augmentation, and management of missing or incomplete EHRs, to support physicians in making quicker and
clinical data to improve model training reliability. During more precise diagnoses. By analyzing large datasets, AI
the training and validation phase, 80% of the dataset was can aid in disease prediction and treatment, helping
allocated for training, with the remaining 20% reserved for clinicians save time and improve patient outcomes. AI
1-3
validation. Model fine-tuning employed backpropagation is instrumental in bringing culturally competent practices
with the Adam optimizer and cross-entropy loss functions to the healthcare industry, ensuring more tailored and
to optimize the performance of the model. Finally, model inclusive patient care. AI’s applications in healthcare span
testing was carried out on unseen data from diverse numerous areas, including diagnostics, treatment protocol
patient cohorts to assess generalizability. Predictions were development, drug discovery, personalized medicine, and
validated against expert-reviewed ground truth labels patient monitoring. In radiology, AI’s role is particularly
and clinical diagnoses, ensuring alignment with clinical noteworthy for interpreting and triaging X-ray images,
standards. A detailed comparative analysis was performed one of the most commonly used imaging tests. AI can
4-6
to assess the AI models across disease categories, focusing analyze these images, helping radiologists prioritize
on the models’ diagnostic efficacy within specific disease critical cases and reducing wait times. However, despite
contexts. In cancer diagnostics, models were evaluated for its promising potential, the widespread adoption of AI
the early-stage detection and tumor segmentation accuracy, in healthcare faces challenges, including ethical concerns
while in cardiovascular diseases, the focus was on detecting about data privacy, job automation, and the amplification
arrhythmias, coronary artery disease, and heart failure. For of biases. Moreover, resistance from healthcare leaders to
neurological disorders, early detection of Alzheimer’s and embrace new AI technologies has slowed its integration
Parkinson’s diseases from magnetic resonance imaging into mainstream medical practices.
(MRI) data was emphasized. For infectious diseases, the In terms of disease diagnosis, AI plays a pivotal role
models’ ability to analyze chest X-rays for the detection of by helping clinicians navigate complex medical data to
coronavirus disease 2019 (COVID-19) and tuberculosis identify conditions accurately. By leveraging vast EHR
was examined. This cross-disease comparison highlighted datasets, AI algorithms can predict diseases such as
the strengths and limitations of AI models within each Alzheimer’s and dementia, providing early diagnosis and
diagnostic scenario, allowing for an understanding of AI’s potentially improving treatment outcomes. In emergency
7-9
capabilities and challenges in a clinical setting.
settings, AI can prioritize urgent cases by providing real-
To address fairness and ethical concerns, the study time data interpretation to assist decision-making, thereby
incorporated bias detection and mitigation strategies. enhancing efficiency and potentially saving lives. Studies
Subgroup analyses were conducted to examine variations in have shown that AI, through platforms like ChatGPT, can

Volume 2 Issue 3 (2025) 46 doi: 10.36922/aih.5173