Alzheimer’s disease (AD) is an irreversible neurodegenerative disease associated with memory loss, decline in cognitive abilities, and changes in behavior. The most important factor that leads to Alzheimer’s disease for an individual is brain atrophy or an aging brain. With no cure for AD, researchers focus heavily on early detection, which allows patients to receive early treatment and therapies to delay the effects of AD. Thus, patients can improve their quality of life, allowing them to maximize time with loved ones and make informed decisions for the future. Throughout the semester, my research, which includes peer-reviewed articles and medical journals, has been oriented around Alzheimer’s Disease detection and artificial intelligence. Specifically, I have looked into AI’s role in improving early detection. I am looking to address the question: Despite the limitations of AI, how has AI improved the early detection of Alzheimer’s disease?

Currently, the overall conversation about AI and medicine has been positive, as medical professionals are open to integrating AI into AD detection. One way AI proves to be beneficial is through the involvement of neuroimaging. In the article, Artificial intelligence in Alzheimer's diagnosis: A comprehensive review of biomarkers, neuroimaging, and machine learning applications, researchers reveal that AI can analyze neuroimaging more quickly and precisely than traditional methods (Cretu 2024). Neuroimaging includes Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), and Computed Tomography (CT). Additionally, traditional methods involve having radiologists sit down and analyze such data, which can lead to human errors and be a time-consuming process compared to AI. Along with analyzing data, the article, Potential Ocular Biomarkers for Early Detection of Alzheimer’s Disease and Their Roles in Artificial Intelligence Studies, reveal that using “AI with ocular biomarkers to detect AD reported promising results, demonstrating that using AI with ocular biomarkers through multimodal imaging could improve the accuracy of identifying AD patients” (Chaitanuwong 2023). Therefore, AI has the potential to open doors for more methods of detection. In the article, Unveiling New Strategies Facilitating the Implementation of Artificial Intelligence in Neuroimaging for the Early Detection of Alzheimer’s Disease, research proves that AI has been beneficial by allowing patients to have better accessibility to detection. The study mentions the use of recording speech audio with smartphones for the detection of AD symptoms (Etekochay 2024). Ultimately, the current conversation in AI with AD detection has been extremely positive.

With the help of AI, early detection has been more advanced than ever before. However, artificial intelligence is not perfect. Some limitations exist within these tools, which researchers must take into account. Currently, some of the limitations of AI in AD detection include the limited access to data that AI models are trained on and the limited explainability that AI models provide. This paper will give an overview of such limitations and explain the current progress of combating these challenges. Ultimately, throughout research, I have concluded that despite limitations, AI still proves to be beneficial in early AD detection.

One of the limitations of AI in AD includes the limited access to data cohorts. However, it is important to understand how AI models function first. AI models have a very specific way of training on data to make informed decisions. First, the model will take in a lot of information from data. In AD detection, AI models will take in data cohorts that would contain data from patients with Alzheimer’s disease. This includes their demographics, cognitive assessments, neuroimaging, biomarkers, and longitudinal data (Alzheimer's Disease Neuroimaging Initiative 2025). Then, AI models will train based on real data by taking in some input, making a prediction, and comparing the prediction to the actual results. Lastly, the model will update itself based on these inputs and predictions to become more and more accurate. Therefore, the more data available means more training, which means more accuracy.

As previously mentioned, AI models can make predictions for a wide range of data, ranging from analyzing imaging data such as MRI, CT, and PET scans, to biomarker data such as ocular biomarkers (Cretu 2024). In the article, ADataViewer: exploring semantically harmonized Alzheimer’s disease cohort datasets, researchers describe data cohorts to have “laid the foundation to discover novel biomarkers, investigate disease progression, and identify disease subtypes” (Salimi 2022). However, the problem that researchers face is that when collecting data to train models, they have to manually validate each cohort before testing. Depending on how many cohorts are in use, this can be a time-consuming process as researchers need to be granted access to data, validate the data, and recategorize variables. Recategorizing variables is difficult, as each cohort would have different naming systems and measurements (Salimi 2022). Clearly, the process of accessing data can hinder advancements by weeks to months at times.

Another limitation with AI models involves the limited explainability when predictions are made. AI models include deep learning (used to learn and predict based on large datasets) and neural networks (which makes decisions similar to a human brain). Typically, AI models will take in some input and give an output without any explanation. This is called the “black box” outline. In Vernekar’s article, Exploration of explainable AI with deep learning model for early detection of Alzheimer's disease, researchers explain that “while deep learning models like CNN and CapsNet achieve high performance in detecting Alzheimer's, they often function as “black boxes,” making it difficult for clinicians to understand how decisions are made. This lack of transparency limits clinical adoption and trust. Existing studies on Alzheimer's detection primarily focus on improving accuracy but overlook explainability” (Vernekar 2024). “Black box” cases leave plenty of room for error and skepticism for professionals, as there is no way of manually verifying such decisions. An analogy to this situation would be answering math problems on a math exam. If all answers were written with zero work shown, graders would be skeptical about where the answers came from. This scenario can be applied to AD detection, too. Therefore, another problem that arises is the lack of explainability in AI.

Innovative solutions are in the process of being implemented to eliminate the issue with extensive data collection. For instance, in 2022, Salimi et al’s research group introduced the platform: ADataViewer. Previously, researchers may have had a hard time accessing different data cohorts. However, this platform “enables the scientific community to explore 20 AD cohort datasets, both from a semantic and statistical perspective.” These cohort datasets include the National Alzheimer’s Coordinating Center (NACC) and Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease (A4). As previously mentioned, each cohort may have a unique variable system, which makes cross-referencing difficult. To combat this issue, ADataViewer presents “a variable mapping catalog that harmonizes 1196 unique variables encountered in the datasets, spanning nine data modalities.” Additionally, researchers often desire diversity and a vast amount of data to ensure accurate training and avoid bias for AI models. Therefore, ADataViewer presents a solution by allowing researchers to “facilitate the exploration of the cohort datasets with respect to longitudinal follow-up, demographics, ethnoracial diversity, measured modalities, and individual variables” (Salimi 2022). This source also reveals that the patient baseline for each dataset ranges from 90 to 40,948 patients. The versatility of this tool greatly minimizes the time researchers spend on collecting data. Furthermore, AI models can train with this data to improve the early detection of Alzheimer’s Disease in a more accurate and nonbias manner. With this solution and further research, the limitation of accessing data can be resolved.

Researchers are in the process of developing solutions to the limitations of explainability in AI models. For instance, AI is heavily used in analyzing neuroimaging data, specifically Magnetic Resonance Imaging (MRI) scans. MRI is a critical tool in identifying brain atrophy, thereby improving the early detection of AD, and helps with monitoring disease progression. Furthermore, MRI scans have played a role in allowing researchers to have a better overall understanding of AD. AI models, specifically deep learning models along with neural networks, will go through MRI data to differentiate patients with healthy brain functions or AD symptoms. However, many models have a “black box” outline, which results in skepticism in clinical trust. In the article, Biomarker Investigation Using Multiple Brain Measures from MRI Through Explainable Artificial Intelligence in Alzheimer’s Disease Classification, Coluzzi et al. developed Explainable Artificial Intelligence (XAI) to combat this challenge. Researchers explain that XAI is a technique that aims “to bridge the gap between the need for high performance and that of understanding the processes leading to the final classification. In this context, no studies i) systematically verify whether the brain regions leveraged by the DL model align with the established knowledge of AD and ii) employ DL in both brain connectomes and neuroimaging data for casting a new light on the neurodegenerative mechanisms underlying the disease” (Coluzzi 2025). In the study, researchers focused on two deep learning models that analyze MRI Data and then paired them with the XAI method. They were able to uncover the ultimate drive for the decision-making process for both models. Ultimately, XAI puts “emphasis on the potential combination of imaging and connectivity data as a means to create better and more reliable AD models” (Coluzzi 2025). By having more explainability models paired with AI models, medical professionals can develop more reliability and trust in models, which leads to more efficient and precise early detection of AD. Furthermore, by intertwining human analysis and AI analysis, fewer errors can occur, which allows for better diagnosis and understanding of the disease.

Through my research, I have concluded that despite limitations, AI is playing a huge role in the early detection of Alzheimer’s disease. Previously, limitations of AI models include the barriers to accessing data and the lack of explainability in AI decision-making. However, transformational initiatives have been developed to combat these challenges. For example, ADataViewer developed by Salimi et al., has been efficient in allowing researchers to access a multitude of data. This data can be used to help deep learning and neural network models with training. Through training, models become more accurate, which allows for the successful detection of AD. Additionally, XAI, along with other explainability models, can be implemented with AI to garner more clinical trust. Therefore, AD detection can have fewer errors and be more time-efficient.

Ultimately, Alzheimer’s Disease is a terrifying disease that has no cure, which is why early detection plays a huge role in improving the quality of life for patients. Through early detection, professionals can help delay the effects of AD and allow patients to manage their symptoms. I believe the most important factor is time. Time before symptoms may worsen, time in which patients can spend with their families before AD takes over. Therefore, artificial intelligence plays a role in lessening the time taken for diagnosis, which ties directly into patient care, granting more time for patients. As artificial intelligence improves beyond its limitations, researchers can maximize the efficiency of early detection, which is one step closer to a cure for Alzheimer’s Disease.

Sources

• Alzheimer's Disease Neuroimaging Initiative. (2025). Alzheimer’s Disease Neuroimaging Initiative. https://adni.loni.usc.edu/
• Chaitanuwong, P., Singhanetr, P., Chainakul, M. et al. Potential Ocular Biomarkers for Early Detection of Alzheimer’s Disease and Their Roles in Artificial Intelligence Studies. Neurol Ther 12, 1517–1532 (2023).
• Coluzzi, Davide et al. Biomarker Investigation Using Multiple Brain Measures from MRI Through Explainable Artificial Intelligence in Alzheimer’s Disease Classification. Bioengineering. 12.1 (2025): 82–109.
• Crețu, O. C., Cocu, M., Popescu, E. R., & Chiriță, R. (2024). Artificial intelligence in Alzheimer's diagnosis.
• Etekochay MO, Amaravadhi AR, González GV, et al. (2024). AI strategies in neuroimaging.
• Salimi, Yasamin et al. (2022). ADataViewer: Exploring Semantically Harmonized Alzheimer’s Disease Cohort Datasets.
• Vernekar, S. R., & Kumar, S. (2024). Explainable AI for early detection of Alzheimer's disease.