Figure 1:
Figure 2:
Figure 3:
Figure 4:
Current challenges in cancer detection using ML
| Aspect | Challenges | Description | Reference |
|---|---|---|---|
| Cancer detection | Lack of high-quality training data | AI models require large, well-annotated datasets, which are often limited for certain cancer subtypes. | [34] |
| Model interpretability (“Black Box” problem) | Complex DL models lack transparency, making clinical validation and trust difficult. | [22] | |
| Heterogeneity in cancer subtypes | Genetic and molecular variations between patients make it difficult to generalize AI predictions. | [57] | |
| Ethical and privacy concerns | Handling patient genomic data poses ethical and legal challenges regarding data security and bias. | [20] |
Cancer toxicity detection methods
| Reference | Description | Advantages | Limitations |
|---|---|---|---|
| [70] | Silico toxicology models present themselves as an alternative to computationally analyze, simulate, and predict the toxicity of novel drugs/compounds. | Obtained the cost-efficiency and Speedy of data. | Not considered the performance-based analysis. |
| [71] | This analysis presented a new model to simulate complex chemical toxicology data sets and compared with the ML methods such as ANN, LDA, SVM, k-NN and NB. | Comparative analysis between the ML models over the drug toxicity. | Not considered the standard evaluation metrics. |
| [74] | Discussed the advances, challenges, and future perspectives of the drug toxicity prediction process using AI. | The various challenges are identified in the process of toxicity prediction. | Future perspective analysis can be enhanced further by considering the latest technology. |
| [81] | This review article discussed in detail about the necessary databases and software that are working based on robust computational assessments and toxicity prediction. | This review provides the analysis of existing databases and software methods. | Not done the robust computational assessments and robust toxicity prediction processes. |
| [82] | Proposed the question about whether clinical trials could be modified/improved for providing the data with better solutions for outstanding issues. | The various issues are discussed by providing the solutions. | The advanced technologies are not considered in the solutions. |
Current challenges in cancer treatment using ML
| Aspect | Challenges | Description | Reference |
|---|---|---|---|
| Cancer treatment | Variability in AI performance | AI-based DSSs have shown inconsistent results in real-world applications, such as IBM Watson. | [21] |
| Computational and resource constraints | Training advanced ML models requires high-performance computing and large datasets. | [50] | |
| Resistance to AI Adoption in oncology | Many oncologists remain skeptical about fully integrating AI-driven decision-making into clinical practice. | [66] |
Current and best methods used in cancer detection using ML
| Method | Description | Advantages | Limitations | Reference |
|---|---|---|---|---|
| ML and DL algorithms | Used for analyzing large datasets, including imaging, genetic, and histopathological data. Helps in predicting tumor recurrence risk and patient responses to therapies. | High accuracy in pattern recognition, automation, and scalability. Can analyze massive datasets efficiently. | Requires large, high-quality datasets. Limited interpretability (“Black Box” problem). | [22, 36, 41] |
| NGS | Enables rapid sequencing of multiple genes, producing vast molecular data for cancer detection. | High throughput, precise genetic mutation detection, and potential for early diagnosis. | High costs, data complexity, and need for specialized expertise. | [28, 29, 43] |
| Radiogenomics | Combines imaging data with genomic data to identify cancer subtypes and predict treatment outcomes. | Non-invasive, links imaging traits with gene expression, and predicts radiation therapy response. | Limited by the availability of high-quality datasets. | [34] |
| Molecular testing | Uses genetic markers such as PD-L1, TMB, MSI, and somatic copy number variations to predict therapy response. | High precision, aids in personalized medicine, and helps in selecting targeted therapies. | Invasive biopsy requirement, limited by tumor region sampling. | [37, 38] |
| Histopathology with AI | AI-powered analysis of stained tissue slides to detect cancer and classify subtypes. | High accuracy, automation, and reduced workload for pathologists. | Requires large labeled datasets for training. | [41, 44, 45] |
| FNA) and cytopathology (TBSRTC) | FNA is a minimally invasive preoperative evaluation, and TBSRTC categorizes thyroid cancer risk. | Quick, minimally invasive, and widely used in clinical settings. | Subjective interpretation can lead to inconsistent results. | [23,24,25,26] |
| Transcriptomics (RNA sequencing, microarrays) | ML-driven analysis of gene expression to classify cancer subtypes and improve diagnosis. | High precision in detecting gene expression differences. | Data complexity and requires advanced computational tools. | [61] |
| Big data and AI integration | AI extracts insights from demographic, clinical, and imaging data to predict prognosis. | Improves decision-making, enhances early detection, and enables better patient stratification. | Underutilized potential and challenges in data integration. | [51,52,53,54] |
| ML-based biomarker discovery | ML helps identify predictive biomarkers, such as PD-L1 expression and TMB, to guide immunotherapy decisions. | Identified the predictive biomarkers | Not considered the suitability level. | [39] |
| Molecular and omics-based analysis | AI integrates genomic, transcriptomic, and proteomic data to improve cancer classification and prognosis. | Performed the effective classification and enriched the cancer diagnosis accuracy. | Not highlighted the false alarm rate and error rate analysis. | [56, 58] |
Current trends in cancer treatment using ML
| Aspect | Current trends | Description | Reference |
|---|---|---|---|
| Cancer treatment | AI-driven personalized therapy | AI models predict individual patient responses to treatments, improving precision medicine. | [35, 40] |
| ML in drug discovery | AI accelerates drug development by predicting the efficacy and toxicity of new compounds. | [50] | |
| CDSS | AI-powered tools assist oncologists in optimizing treatment strategies based on patient-specific data. | [65] | |
| Dynamic dose adjustment with ML | ML models personalize drug dosages based on continuous monitoring of patient responses. | [40] |