Included studies were original research focused on nanomedicine, machine learning, deep learning, and brain disorders. Below, we further describe the eligibility criteria for the included studies.

Population characteristics: Studies involving children and adults, with no restriction on gender, age, race, or ethnicity, were included. Studies that did not involve human participants, such as those utilizing murine models or molecular simulations, were also included.

Intervention/Exposure: Studies were included that show the application of machine and deep learning techniques in advancements of nanomedicine, theragnostic nanomedicine, and photothermal theragnostic for the management of Brain disorders (Brain cancer, Multiple sclerosis, Parkinson’s, and Alzheimer’s disease), but traditional, alternative, or complementary therapy-related studies are excluded. Computational modeling studies, including docking and molecular simulations, were incorporated if there was a direct relationship between machine learning/deep learning and nanomedicine.

Outcome: Studies that examined improving nanomedicine applications such as diagnosis, drug delivery, treatment, prognosis, disease monitoring, biomarker identification and diagnosis, and computational and methodological development by using machine learning, deep learning, or artificial intelligence were included.

Study Characteristics and Design: We included studies from all settings, such as communities, hospitals, specific healthcare facilities, and geographic locations. We included peer-reviewed original research articles.

The articles or studies are excluded if they were editorials, case reports, or interventions treatments that did not involve nanomedicine such as traditional or alternative or complementary therapies. Additionally, conference abstracts and dissertations, grey literature and non-English language articles were excluded from evidence synthesis.

The proper management of brain disorders depends on identifications of early biomarkers and diagnosis for timely interventions. The recent advancements in AI based nanomedicines revolutionize biomarker discovery and enable early diagnosis at a rapid pace and enable personalized healthcare for brain disorders. Surprisingly, there are no reliable biomarkers for Parkinson’s disease (PD) related cognitive decline or dementia. In this context the study done by Chung et al., demonstrated that plasma derived extracellular vesicle (EVs)-borne tau and β-amyloid have the potential to be used as a biomarker for Parkinson’s disease (PD). In this study, researchers isolated EVs from the blood (plasma) of PD patients of varying severity of disease ranging from mild to moderate stages and from control individuals and performed immunomagnetic reduction-based immunoassay to quantify the level of α-synuclein, tau, and Aβ1-42 proteins. Further, patient’s datasets considering attributes such as age, gender and differential expression of these EV markers were used to train the artificial neural network (ANN) model. Interestingly, this supervised model can predict the cognitive dysfunction related to Parkinson’s disease with an accuracy of 91.3%. Consequently, blood-plasma derived EV tau and Aβ1-42 emerged as important predictive biomarkers for early diagnosis and monitoring of PD derived cognitive decline or dementia (43) (Table 1).

Similarly, the conventional diagnostic methods lack early detection of Alzheimer’s disease (AD) because of the absence of reliable biomarkers. In this context, Resmi et al., proposed solution for early detection of Alzheimer’s Disease (AD) using an ultrasensitive SERS-based immunoassay integrated with machine learning approach. This technique utilized an aluminum SERS substrate to detect key AD based biomarkers such as Aβ40, Aβ42, p-tau and t-tau in the blood plasma with sensitivity of attomolar levels. Next, the author used these biomarkers-based datasets and feeds to different machine learning models (MLP, Radial Basis Function, SVM, LDA) to classify and differentiate between mild cognitive impairment (MCI), AD patients, and healthy individuals (44). In conclusion, the development of this AI integrated SERS technique enables the early diagnosis of Alzheimer’s disease by detecting these biomarkers with high precision and accuracy. Like the previous report, Yu et al., suggest that combining SERS with deep learning model (CNN) to analyze cerebrospinal fluid to identify new biomarkers for Alzheimer’s disease (AD). The CNN models are used to analyze CSF based datasets and they can distinguish AD patients with healthy individuals with an overall performance of 92% while 100% for normal individuals, and 88.9% for AD patients. Interestingly, The CNN based classifications are strongly corelated with the clinical dementia rating (CDR) (45). Therefore, this new hybrid technique (SERS and CNN) has potential to identify biomarkers of Alzheimer’s disease with high precision and accuracy detection (Table 1). Thus, the integration of AI with nanotechnology is promising development for the discovery of new biomarkers associated with brain disorders, which is critical for early diagnosis and timely intervention for better disease outcomes.

Furthermore, the integration of AI-inspired machine and deep learning techniques with nanomedicines transforms the diagnosis of neurodegenerative diseases like Alzheimer’s and Parkinson’s, multiple sclerosis and brain cancer. Importantly, AI-based models can be trained by feeding large clinical datasets and these models have potential to analyze large datasets, identifying patterns, and perform extraction of these patterns to make predictions leading to earlier and more accurate diagnoses, crucial for timely interventions and improved patient outcomes. For instance, Corbo et al., developed a non-invasive diagnostic platform for the early detection of Alzheimer’s disease. Notably, this platform utilized a multi-nanoparticle protein corona signature is used to detect a minute change in human plasma protein signatures. Further, these datasets are analyzed by machine learning model including random forest to distinguished plasma samples of healthy and disease subjects with >92% specificity and ≈100% sensitivity, demonstrating its potential for of Alzheimer’s disease (46). Similarly, Etxebarria-Elezgarai et al., used Surface-enhanced Raman Spectroscopy (SERS) with gold nanoparticles (AuNPs) to analyze cerebrospinal fluid samples from Alzheimer’s disease (AD) patients. Combining SERS with AI-based partial least square discriminant analysis (PLS-DA) achieved 100% accuracy in classifying early-stage AD patients compared to 85% for the healthy control group (47) (Table 1).

Importantly, brain amyloid deposition is the hallmark of Alzheimer’s disease, which can be used for diagnosis of Alzheimer’s disease (AD). A newly developed aptamer-based plasma test identifies brain amyloid deposition and, when integrated with a machine learning model, the ExtraTreeClassifier improved the accuracy and predictive power with sensitivity of 0.88, specificity of 0.76 and AUC of 0.79. the overall reports suggest that this method is capable of diagnosis and monitoring the progression of Alzheimer’s disease (48). Interestingly, another scientific group utilized a modified SERS technique with the combination of multivariate statistical methods for the diagnosis of Alzheimer’s disease by analyzing blood serum. In this method researchers used colloidal silver nanoparticles (Ag NPs) as an active SERS substrate.

Furthermore, artificial neural networks (ANNs) were used to analyze SERS spectra to differentiate Alzheimer’s disease (AD) subjects from healthy control in binary model with sensitivity of ≈ 96% and achieved diagnostic sensitivity of 98% for differentiating Alzheimer’s disease (AD) individuals, Healthy control (HC), and Other Dementia (OD) samples in a tertiary model. This study highlights that combination of SERS and ANN models has potential for accurate and early Alzheimer’s disease detection (49). Also, another group used similar concept for the development of modified SERS technique, which involves, combining SERS with AI-based models to enhance detection of Aβ 1–42 monomers, which is important pathological biomarker of Alzheimer’s disease. This SERS platform contains graphene oxide/gold nanohybrids (GO/AuNPs) for the detection of Aβ 1–42 monomers and fibrils form with detection limits of 0.0232 and 0.0192 ng mL⁻¹, respectively.

Next, scientist applied support vector machine (SVM) and one-dimensional convolutional neural network (1D-CNN) algorithms to analyze fibril orientation for accurate diagnosis of Alzheimer’s disease (50). Conventional colorimetric lateral flow assay (LFA) utilizes colloidal gold nanoparticles (AuNPs) for the diagnosis of Alzheimer’s disease, which exhibits low specificity and sensitivity. Importantly, the study led by Wang et al., proposed an advanced method that exploits the machine learning algorithms with optimized colorimetric LFA, and ultrasound enrichment led to detection of tau proteins in undiluted blood serum samples with improved sensitivity. Various machine learning algorithms were used for different purposes such as k-nearest neighbor (KNN) for classification as well as Gaussian process regression (GPR) for accurate quantification of tau protein with enhanced classification, prediction accuracy of 98.11 and 99.99%, respectively, and a limit-of-detection (LOD) of 10.30 pg. mL⁻¹. Thus, this improved method has potential to detect tau protein for clinical diagnosis of Alzheimer’s disease (AD) with greater sensitivity and precision (51) (Table 1).

Additionally, another research group advanced methods involving fluorescent sensor arrays through the integration of machine learning based algorithms including linear discriminant analysis (LDA) for the diagnosis of Alzheimer’s disease (AD). Notably, the sensor array consists of pyrene modified fifth generation poly-amidoamine (G5-PAMAM) dendrimers that allow parallel detection of 11 Aβ40/Aβ42 aggregates including monomer, oligomer and fibril forms present in different interferants, serum media and cerebrospinal fluid (CSF) with enhanced accuracy of 100%. The integration of LDA with fluorescent sensor array has tremendous potential for the early Alzheimer’s disease diagnosis with greater accuracy and precision (52). Similarly, Xu et al., developed another fluorescent sensor array with dual coupling of nonoenzyme (AuNPs) and bio enzyme (horseradish peroxidase) for the detection of ultralow β-amyloid (Aβ) proteins (monomer, oligomer and fibril form) in blood plasma of Alzheimer’s induced mouse model. Further, machine learning algorithms such as k-nearest neighbor (KNN) and linear discriminant analysis (LDA) were employed to identify various β-amyloids aggregates and can distinguish Alzheimer’s induced mouse model and healthy mice with 100% accuracy (53).

The following section explores how the recent synergy of AI driven methods with nano-diagnostic techniques led to the advancements in the diagnosis of Parkinson’s disease. Recently, Xu et al., developed a non-invasive diagnostic method using human saliva samples analyzed with deep learning-based nanoparticle-enhanced laser desorption–ionization mass spectrometry. This approach successfully mapped the saliva metabolic fingerprint helps to differentiate between healthy subjects and Parkinson’s disease participants with an AUC of 0.8496 (54). Moreover, deep brain stimulation (DBS) has been used to treat Parkinson’s disease, but the traditional use of lead electrode poses a risk of metal toxicity. To address this, scientist have developed a new approach called Lead-free piezoelectric nanoparticle-based DBS (LF-PND-DBS). This innovative method utilizes lead free piezoelectric nanoparticle to deliver electrical stimulation and thus eliminating the risk of metal toxicity (Table 1).

Recently, Eid et al., further advanced this technology by incorporating machine learning techniques such as transformer networks, hybrid simulated annealing–particle swarm optimization (SA-PSO), federated learning using an optimized LF-PND-DBS system. In a cohort patients study this optimized system achieved high accuracy (99.1%) and specificity (98.2%) for diagnosis and demonstrating its potential for precise diagnosis and personalized treatment delivery (55). Furthermore, another interesting study done by Li et al., developed a non-invasive method for diagnosing Parkinson’s disease using plasma blood samples. The authors have analyzed small molecule miRNA associated with extracellular vesicles (EVs) in the plasma of healthy and diseased patients by applying support vector machine (SVM) model. This algorithm can effectively identify EVs miRNA signatures and has potential to distinguish between plasma samples from healthy control and those with Parkinson’s disease individuals with an AUC of 0.916 (56). In conclusion the integration of AI driven techniques with nano-diagnostic tools substantially improved the accuracy, specificity, sensitivity, and predictive abilities leading to precise classification and early detection of the Alzheimer’s and Parkinson’s diseases pave the way for timely intervention and improved patient outcomes (Table 1).

Multiple sclerosis is a chronic neurodegenerative disease and it is conventionally diagnosed with the help of invasive methods. To overcome these challenges researchers are actively pursuing non-invasive solutions, particularly through the integration of AI and nano-diagnostic techniques. In this context, the study led by Broza et al., developed a rapid, non-invasive technique analyzing volatile organic compounds (VOCs) in breath samples using gas chromatography–mass spectrometry (GC–MS) and a nanomaterial-based sensor array and built a predictive model with an artificial neural network (ANN). GC–MS was used for analyzing distinct VOC profiles in MS patients compared to controls. The combination of sensor array and artificial neural network can easily differentiate between MS patients from heathy control group with accuracy of 90%. Also, the blind validation was carried out and it can differentiate other studied groups with great accuracy such as 95% PPV for MS remission versus control group with 100% sensitivity and 100% NPV for MS non-treated versus control groups, and 86% NPV for relapse versus controls (57). In conclusion, AI powered GC–MS methods have potential to diagnose multiple sclerosis with improved accuracy (Table 1).

Furthermore, the integration of AI based models with core techniques utilizing nanoparticle or nanomaterial has improved the diagnosis of brain cancer. In this context, Eyruad et al., utilized nano differential scanning fluorimetry (nanoDSF) to generate plasma denaturation profiles (PDPs) from the blood plasma of brain cancer patients and healthy individual’s subjects. PDPs are further analyzed by AI based algorithms to detect altered EGFR expression profile and perform automated classification with accuracy of 81.5% (58). Thus, this technique can be useful for the better diagnosis of brain cancer associated with alteration of epidermal growth factor receptor (EGFR). Importantly, Rani et al., utilized nanoscale imaging with deep neural network-based segmentation to locate tumor in precise manner in MRI based images, which cannot be possible by using conventional methods. The integration of neural networks with MRI can easily locate tumors with a high accuracy of 97.3% (59). Moreover, the complete surgical removal of brain tumors is essential to prevent the risk of glioma recurrence, but the invasive nature of tumor cells makes surgery challenging due to limitations in identifying tumor boundaries or margins.

To address this limitation Sun et al., introduced modified surface-enhanced Raman scattering (SERS) technique for detection of glioma cells using silver nanoparticles (AgNPs). Further, the SERS produced spectral peaks are analyzed using artificial neural networks and polynomial regression modeling. The peak around 655 and 717 cm⁻¹, correspond to glioma cell proportion and can estimate the glioma cell percentages in simulated sample and frozen sample with R² of 0.98 and 0.85, respectively. Therefore, the integration of AI based algorithms with SERS would be useful for identification of tumor margins in fresh tissue samples and it could be helpful for real time visualization of tumor boundaries during intraoperative brain tumor surgery (60).

Similarly, Wang et al., introduced the concept of differentiating glioma at both cellular and tissue levels. To make a distinction at the cellular level between normal astrocytes and non-central nervous system (CNS) tumor cells, the researchers used a modified SERS technique utilizing gold nanoshell (SiO₂@Au) particles and support vector machine (SVM) algorithms. Subsequently, gold nanoisland (AuNI) SERS substrates used for detection of glioma tumor tissue and further employing support vector machine (SVM) and orthogonal partial least squares discriminant analysis (OPLS-DA) to distinguish glioma and trauma tissues, identification of various tumor grades, and determination of IDH mutation with great accuracy and precision. These findings indicate the feasibility of SERS-based methods for both single-cell and tissue-level GBM detection, with potential for real-time intraoperative diagnosis (61). In conclusion the integration of AI driven techniques with nano diagnostic tools significantly improved the accuracy, predictive abilities leading to precise classification and detection of the glioma Versus normal cells demonstrating the revolution of the field of brain cancer diagnosis (Table 1).

The management of brain related disorders also depends on prognosis and monitoring the severity of disease in regular intervals of time. Currently, the recent developments and progress in AI based nanomaterials are used to track the disease by monitoring the expression level of specific biomarkers, neurotransmitters release, and analyzing image datasets (Table 2). This section will explore how different nanomaterials and AI based approaches are used for prognosis and disease monitoring of brain related pathologies.

In both Alzheimer’s and Parkinson’s diseases, it is evident that the accumulation of oligomers plays a critical role in the neurodegeneration process. Aβ and α-synuclein are the key proteins involved in these conditions. Therefore, it is essential to develop high-throughput methods for quantitatively estimating these oligomers with high accuracy and precision. To address this problem Sandler et al., proposed a method that combined solid-state nanopores and DNA barcoding to detect and quantify α-synuclein oligomers. Additionally, researchers developed a multiplexed detection system to detect and analyze multiple samples or different conditions within a single experiment. The nanopore sensing detection systems measured change in ionic current based on the analyzed proteins size, shape, and charge. Also, it allows the testing of drugs that inhibit oligomer agglomeration. Researchers employed the combination of molecular docking and iterative active learning to identify small molecule inhibitors for α-synuclein aggregation. Furthermore, these potential drugs were identified using machine learning models that continuously updated and refined their predictions as new experimental data were generated. As a result of this the two potential inhibitors (Anle-138b and I3.08) were identified, which can inhibit α-synuclein’s secondary nucleation (62) (Table 2).

Another strategy for studying the aggregation of these oligomers is to develop methods capable of early detection of this process. Tahirbegi et al., proposed a single-molecule photobleaching (SMP) technique to detect early-stage aggregation of Aβ40 and α-synuclein. This method outperforms conventional techniques that struggle to identify the transition from monomer to low-order oligomer at low physiological protein concentrations. Typically, proteins are immobilized on a solid surface to ensure single-molecule resolution. In the case of α-synuclein samples, these samples were collected after 12 h of incubation with various concentrations of gold nanoparticles. These nanoparticles function as aggregation modulators, exhibiting inhibitory and promotional effects depending on their concentration. Subsequently, Total Internal Reflection Fluorescence microscopy recorded fluorescence intensity over time. The data obtained from the fluorescence was used to train SVM and MLP models. Overall, the MLP model achieved an accuracy of 83.5%, with a relaxed true positive rate (TPR) of 98.9% (63). The AI-driven advancements in nano diagnosis methods enable the potential to monitor the progression of Alzheimer’s and Parkinson’s diseases (Table 2).

The study, by Chan et al., utilized a peptide-encapsulated droplet micro laser and a 3D deep learning strategy such as multi-convolution architectures (2D + 1D), along with fully connected layers to detect minute nano structural spectral shifts during amyloidogenesis and to monitor their progression. Interestingly, this multimodal approach achieved high accuracy of over 95% for all kinds of datasets including training, validation, and test sets and demonstrating AI based nanomaterials have the potential to its potential for the progression of these neurodegenerative diseases (64). Furthermore, the integration of AI and nanotechnology enables us to develop strategies to detect and track the changes in neurotransmitters. In this context, Hozhabr et al., developed a multicolor sensor using gold nanorods to detect and monitor dopaminergic drugs (L-DOPA, carbidopa, benserazide). Researchers employed the linear discriminant analysis (LDA) and partial least squares regression (PLSR) to process high-dimensional data for enhanced detection with achieved 100% accuracy in classification and quantification tasks and detection limits ranging from 0.03 micromolL⁻¹ to 0.9 micromolL⁻¹ (65).

Moreover, other methods have integrated the detection of neurotransmitters in brain tissue which is demonstrated by Komoto et al., who report that nanogaps and XGBoost are used to measure tunneling currents to detect specific neurotransmitters. XGBoost served as a qualifier for differentiating neurotransmitters. Although the metrics were not satisfactory (F1-score 0.52), the single-molecule measurement enabled high-resolution monoamine neurotransmitter detection in both solutions and mouse brain tissue (66). One type of nanoscale biomarker is extracellular vesicles (EVs), which are implicated in Alzheimer’s and other tauopathies. These vesicles have been found to transport tau and amyloid-beta oligomers between brain cells. In the Muraoke et al., study, plasma samples were collected from NFL players and age-matched controls to isolate EVs using size-exclusion chromatography, followed by proteomics analysis with mass spectrometry. The study also evaluated total tau (t-tau) and phosphorylated tau (p-tau181) levels using a modified, ultrasensitive immunoassay. These protein levels were used to train models, including linear discriminant analysis, Naive Bayes, and SVM, to differentiate between NFL players and the control group with achieved accuracy of 85% and AUC of 0.85. The finding suggests that these EVs contained tau proteins are associated with the progression of Alzheimer’s disease (67) (Table 2).

Besides detecting proteins, imaging is also utilized to track the biomarkers in neurodegenerative diseases such as multiple sclerosis and Parkinson’s. Similar to this concept, Crimi et al., introduced machine learning-based classification framework to identify spatiotemporal patterns of lesions observed in multiple sclerosis. In this study, the author used ultrasmall superparamagnetic iron oxide (USPIO)-enhanced MRI to highlight the macrophages activity. Further different clustering methods such as K-means, hierarchical clustering, Gaussian mixture models (GMM), and spectral clustering were applied for a two-tier classification framework, including lesion classification and for patient classification.

Furthermore, scientists developed a tremor sensor to monitor the Parkinson’s disease (PD) severity and treatment response by tracking tremor patterns of patients in a real-time manner (68). The study led by Kim et al., worked on this concept and introduced a stretchable and self-healable catechol-chitosan-diatom hydrogel (CCDHG) as a novel biocompatible electrode for triboelectric nanogenerators (TENGs) which can be utilized as a tremor sensor to detect low-frequency vibrations in PD patients. Additionally, scientist applied machine learning algorithms such as linear SVM and KNN that enables sensor to classify the tremors based on different tremor intensities with an accuracy of 100% (69). Similarly, monitoring disease progression of multiple sclerosis by employing nanomaterial is reported by Morris et al., in this study authors implanted microporous poly (ε-caprolactone) (PCL) scaffolds subcutaneously in SJL/J mice to examine the immune cell activity and gene expression pattern at different stages of the disease (pre-symptomatic, symptomatic, remission, relapse). Further, a Bagged tree algorithm was used to predict the disease state and relapse by analyzing changes in gene signatures. This minimally invasive method allows us to study both the disease’s mechanism and the responses to treatments in MS and other immune diseases (70) (Table 2).

This section explores the recent development in AI based nanomedicine led to transformational changes in drug delivery methods and treatment strategies for the management of brain disorders. Also, covers different machine and deep learning strategies that improves pattern recognition and extraction from the different datasets.

The conventional machine learning approaches used input features as a static value, but the study done by Munteanu et al., presents a novel approach using perturbation theory machine learning (PTML), which dynamically adjusts molecular descriptors, such as logP, by incorporating perturbations observed across different experimental conditions (e.g., cell types) and then uses these values as input to train the ML model. The authors used molecular descriptors of the drug, the nanoparticles, experimental conditions, and the perturbations to which the system is exposed. Despite testing approximately 800,000 drug-nanoparticle complexes, no pair of clinical interest is mentioned; however, among the relevant features that were found, given that the best model was a decision tree-based method, are Perturbation of Polar Surface Area in different cell types, perturbation of logP (lipophilicity) in other cells, nanoparticle surface area of acceptor atoms, nanoparticle size in various experimental conditions, Van der Waals volume of nanoparticles in different situations, and nanoparticle polarizability in other conditions. This shows that drug solubility and nanoparticle physicochemical properties in predicting cytotoxicity are essential in predicting a cytotoxic complex and collaborate in the design of new treatments (71) (Table 3).

Additionally, advances in glioblastoma and glioma have increasingly focused on identifying specific molecular targets for therapeutic intervention. The epidermal growth factor receptor, or EGFR, is a tyrosine kinase receptor widely studied in cancer. The mutation in EGFR gene is responsible for dysregulated protein expression results in abnormal cellular functions including survival, proliferation, differentiation and migration. In this context, Chandra Kaushik et al., conducted in silico study that applied deep learning, molecular docking, time course simulation and synthetic biology to identify the nanoparticle which has synergistic antitumor activity with anti-EGFR-iRGD protein. In this study, author employed deep learning models were used to predict potential nanoparticle which has higher binding affinity with maximum inhibitory potential against EGFR-iRGD protein using PubChem chemical compound library. The predictions suggest that Gold Nanoparticles (AuNPs) were the most effective inhibitors of EGFR. Further, to validate this finding researcher performed molecular docking study to determine the binding affinity and found that AuNPs and EGFR have a binding affinity of (−3.5 kcal/mol). Additionally, the authors applied synthetic biology to examine the effects of EGFR and AuNPs interaction in downstream signaling pathways including Ras/Raf/mitogen-activated protein kinase pathway, the PI3K/Akt pathway, the PLCγ pathway, and the STAT3 pathway by applying Boolean circuit analysis. Through this analysis, the researchers validated the impact of AuNPs (0.45 mmol) on EGFR downstream signaling pathways over time, and found that AuNPs gradually diminish EGFR activation, which results in inactivation of downstream players leads to tumor suppression. Overall, results suggest that antitumor efficiency of anti-EGFR-iRGD protein with gold nanoparticles (AuNPs) complex is effective against EGFR driven tumors (72). Thus, this study opened the door to predicting and validating nanoparticle-based therapies that target specific downstream molecular pathways, providing a new basis for developing effective precision medicine (Table 3).

Moreover, there is a significant advancements in imaging techniques are also reported, which is evident from the study conducted by Karthik et al., who introduced a hybrid CNN-GAN model to improve tumor segmentation and refine radiotherapy targeting through quantum dots (QDs). In this approach the quantum dots provide high-resolution contrast at varying tissue depths, thereby improving the performance of real-time MRI while CNN-GAN hybrid model substantially improved the extraction and segmentation of brain tumors. CNN contributes to feature identification by recognizing distinct patterns in healthy and tumor tissue while determining characteristics such as size, shape, and heterogeneity. Meanwhile, GAN utilizes a generator-discriminator framework that refines the segmented images to capture specific tumor details. The researchers used high-resolution 3D MRI brain scans to segment the tumors and compare their findings with the ground truth, achieving a Dice Coefficient of 0.95, which confirms high segmentation precision. The results indicate a notable advancement in neuro-oncology, as it allows for the proposal of real-time and accurate patient treatment (73) (Table 3).

While nanomedicine and AI have made significant strides in cancer treatment by optimizing drug delivery and imaging precision, similar approaches have also been directed toward Parkinson’s and Alzheimer’s diseases. This field focuses on enhancing drug bioavailability, penetrating the blood–brain barrier, and targeting neuromodulation. Specifically, machine learning and deep learning models are utilized to improve drug treatments and brain stimulation. In the realm of neuromodulation, as discussed in the Diagnostic/Biomarkers section, Eid et al., proposed integrating machine learning methods to diagnose and evaluate treatment efficacy using lead-free piezoelectric nanoparticle-based deep brain stimulation (LF-PND-DBS). They trained machine learning models using EEG datasets from Kaggle and clinical records of Parkinson’s to optimize DBS stimulation parameters. In this instance, the best-performing model was based on transformers hybridized with Simulated Annealing and Particle Swarm Optimization, producing very promising metrics; for example, the F1 score achieved a value of 99.2. This study paves the way for advancements in neuromodulation without requiring invasive techniques, minimizing adverse effects, and integrating personalized treatment (55) (Table 3).

Among the treatments for the symptoms of Parkinson’s disease, there are also drugs such as selegiline HCl, which works alongside levodopa to slow the progression of Parkinson’s symptoms. Many drugs have difficulty crossing the blood–brain barrier (BBB), making it essential to find different dosage forms that enhance drug bioavailability. In this context, Kakulade et al., developed, characterized and evaluated the pharmacokinetics of selegiline HCl (SGH)-loaded cubosomal thermoreversible mucoadhesive gel for nose-to-brain drug delivery. The formulations utilized cubosomal gel as a drug delivery vehicle which can transpassing the blood brain barrier (BBB) and carrying both hydrophobic and hydrophilic drug. In this study, artificial neural network was used to optimize formulation by considering various parameters including formulation variables (glycerol monooleate, Poloxamer 407, Tween 80) and output variables (particle size, entrapment efficiency (EE), and drug release). The study showed this AI based approach optimized nanoparticle size with controlled drug release for up to 6 h, and a cumulative drug delivery rate in the cubosomes exceeding 70% with 1.90-fold increase in pharmacokinetic performance in brain compared to the drug solution alone (74). Similar study done by Sun et al., explore the effect of molybdenum disulfide (MoS₂) nanosheets and MoS₂ crystal structure on cellular functions including adhesion, differentiation, and neuroprotection. These findings revealed that the octahedral conformation of 1 T-MoS₂ nanosheets can to regulate biomolecular interactions, resulting in enhanced neuroprotection. Further, researchers employed random forest analysis to determine the important variables to quantify the relationship between the nano structural architectures. Lastly, structural equation modeling (SEM) confirmed a causal relationship, showing that neurite outgrowth directly related with increased cell survival. Finally, the researchers discovered that MoS₂ laminas can alter the folding of alpha-synuclein, thus preventing its aggregation. Molecular dynamics simulations showed that 1 T-MoS₂ interacts with α-synuclein fibrils, destabilizing β-sheet structures and facilitating fibril disaggregation. In a Parkinson’s disease mouse model, direct striatal injection of MoS₂ nanosheets enhanced memory retention and reduced neuroinflammation. Therefore, this nanosheet has therapeutic potential against Parkinson’s disease (75) (Table 3).

Applying machine learning and deep learning models in nanomedicine permeates different fields beyond the clinical area. Advances in brain disorders include the nanoscale study of other molecular components, image analysis in the study of tissues, and even molecular interactions. By combining AI and nanomedicine, researchers are overcoming traditional methods, challenges and limitations, offering more accurate and cost-effective solutions. This section explores recent computational advances in nanoparticle interactions, imaging analysis, and molecular sequencing. As we have observed up to this point, integrating AI and nanotechnology drives throughput clinical precision, increases data granularity, and improves diagnostic accuracy. AI methods’ ability to extract patterns from data, be it images or tabular data, allows processing of large amounts of data and increases temporal and spatial resolution.

In the case of brain tumors, these advancements also allow for the segmentation of intricate tissue architectures and the integration of multimodal data to improve diagnostic accuracy. One of the most pressing challenges in brain tumor research is understanding intra-tumoral heterogeneity (ITH), the dynamic and spatially variable nature of tumor populations that evolve rapidly and respond differently to treatment. Traditional biopsy-based molecular characterization tests are biased by sampling bias, missing critical regions of genetic diversity. Thus, creating a non-invasive method to characterize ITH across space and time is even more challenging. The Parker et al., study proposed a multimodal integration method of multiparametric magnetic resonance to map tumor properties across multiple spatial scales, from macro to micro and nano levels. The idea was to decode molecular and cellular behaviors from macro scales as provided by MRI. For this, incorporating genetic markers (e.g., IDH1 mutation, MGMT promoter methylation) with imaging features (e.g., signal intensity, diffusion metrics, perfusion parameters) was proposed, which provided a more accurate and dynamic measurement of tumor heterogeneity (76). The proposed model is a good example of the synergy between deep learning models; the authors propose to use convolutional neural networks for feature extraction from MRI images for tumor segmentation, while with generative adversarial networks, researchers can perform data augmentation, improve the quality of the images, and improve the generalization of the model. Beyond deep learning, a Random Forest was used to classify tumors based on molecular and imaging features. Support Vector Machine was used to identify the most informative imaging features linked to molecular subtypes. Although this study primarily serves as a roadmap review, the authors propose a pipeline that allows predictions to help classify treatment-resistant tumor subpopulations.

Additionally, drug perfusion in brain tumors is often very complex due to irregular vascularity and uneven passage through the BBB. Current methods, such as transcardial perfusion, cannot accurately determine drug extravasation because many remain in the vasculature, leading to potentially overestimated measurements. The study by Kostrikov et al., 3D deep tissue imaging with optical tissue clearing was used to capture three-dimensional images of the tumor vasculature. TRITC-dextran, a fluorescent marker that permeates tissues but not cells, was utilized to investigate extravasation. Further, the CNN model (VGG-19) was used for feature extraction or specific patterns in the tumor images while random forest was applied for classification particularly to identify extravasation points in the images. Moreover, to get better insight into the drug distribution within the tumor tissue, the researchers calculated the surface area and volume of the extravasation zones based on the number of identified spots and interestingly, found that the surface area and volume of extravasation zones were greater in more permeable areas of the tumor vasculature (77) (Table 4).

Apart from image-based analysis, molecular interactions are also studied at nanoscale resolutions. The study done by Liu et al., suggests that the catalytic activity of acetylcholinesterase (AChE) enzymes is inhibited by surface-functionalized gold nanoparticles (f-GNPs). This inhibition results in increased acetylcholine (ACh) release, which is responsible for neuroprotection against cholinergic neuron death in Alzheimer’s disease. In this study, a set of 47 f-GNPs was screened and experimental analysis showed that allyl-functionalized nanoparticles with fewer Hydroxyl group (-OH) showed maximum inhibition efficiency. Moreover, researchers trained a Bayesian neural network model to enhance the results and predict AChE inhibition by the f-GNPs. Thirteen molecular descriptors derived from quantitative structure–property relationship (QSPR) analysis were used as training data. Since nanoparticles can interact with AChE through various mechanisms including specific and nonspecific binding, lipophilic interactions, hydrogen bonding, and charge/dipole interactions these studies help identify the binding affinity and inhibition potential of different nanoparticle designs (78) (Table 4). Consequently, these results represent a significant advance in screening nanomaterials for treating cognitive disorders in Alzheimer’s disease, as they enable early prediction of nanoparticle efficacy and toxicity before synthesis. In the same line as the previous research, Zhang et al., developed a copper (II)-functionalized Mycobacterium smegmatis porin A (MspA) nanopore to interact with proteins; they intended to generate a detection method of 20 proteinogenic amino acids. The authors used machine learning methods for the classifier, including Random Forest, Naïve Bayes, Neural Networks, and Ensemble methods as input data, they used blockade signals of amino acids passing through the nanopore and achieved a 99.1% classification accuracy with a 30.9% signal recovery rate. Also, this method is utilized for peptide sequencing associated with Alzheimer’s disease and cancer. Importantly, the sequencing of amyloid-beta (Aβ) peptides could be performed through this approach which is an important biomarker. Importantly, this new approach has detection limit in nanoscale which is less than 100 nM. It has potential to identified all amino acids, two post-translational modifications (PTMs), and one unnatural amino acid with 99.1% accuracy. Although this method is applicable for peptide sequencing but not for whole proteins sequencing. Thus, it offers a promising pathway toward nanopore-based proteomics (79) (Table 4).