AI Moves From Screening Aid to Daily Decision Support in Lung Cancer Care
Artificial intelligence is moving into lung cancer care at multiple points along the clinical pathway, from screening and diagnosis to treatment selection and follow-up. In thoracic oncology, where small changes in histology, molecular profile, or treatment response can alter the entire course of care, the appeal is obvious: better pattern recognition, faster processing of complex datasets, and more consistent interpretation of imaging and pathology. The review frames AI not as a single tool but as a collection of methods that can be applied to distinct problems in precision oncology.
The central promise is operational as much as biological. Lung cancer care generates enormous volumes of data, including CT images, pathology slides, genomic profiles, treatment records, and longitudinal toxicity assessments. Human teams already use all of these inputs, but not always with the speed or scale needed to extract every actionable clue. AI systems are designed to find recurrent structures in those datasets, which may help identify malignancy earlier, refine risk stratification, and support treatment planning when the margin between benefit and harm is narrow.
One sentence captures the clinical tension: AI can process the data faster than any tumor board, but it still cannot replace the judgment required at the bedside.
Screening, Imaging, and Pathology Are the First Entry Points
In cancer screening, AI has emerged as a strategy to improve diagnostic accuracy, efficiency, and scalability. For lung cancer, where early detection remains one of the most important determinants of outcome, that matters immediately. Algorithms built on imaging data can be trained to detect subtle radiographic patterns that may be difficult to appreciate on a single read, particularly when nodules are small, indeterminate, or embedded in complex underlying lung disease. The practical value lies in triage: flagging suspicious lesions sooner, reducing missed lesions, and helping radiology services manage increasingly large screening volumes.
Pathology offers a second major frontier. Deep learning methods applied to whole-slide images, often described as pathomics, can identify features that correlate with histology and molecular status. Those features may have direct relevance to treatment selection because many decisions in thoracic oncology now depend on biomarkers rather than anatomy alone. If an algorithm can infer likely molecular characteristics from tissue architecture or image-based features, it may one day help prioritize samples for further testing or point clinicians toward a targeted strategy earlier in the workflow. The same logic extends to radiomics, where quantitative imaging features may help predict biologic behavior before surgery, biopsy, or a treatment switch.
Molecular Prediction and Drug Discovery Could Shorten the Distance to the Right Therapy
The review also places AI in drug discovery, where computational methods can interrogate large genomic, chemical, and clinical datasets to identify targets and match candidate compounds to available vulnerabilities. In lung cancer, this is especially relevant because resistance to targeted therapy remains a major clinical barrier. Tumors evolve. Molecular escape mechanisms emerge under treatment pressure, and the next effective therapy often depends on recognizing those changes quickly enough to act.
AI is well suited to that problem because it can integrate diverse signals that do not always sit neatly in the same chart. Genomic data may suggest a driver alteration. Chemical data may point to a candidate compound. Clinical outcomes data may indicate which combinations are tolerable, durable, or prone to failure in the real world. A machine-learning framework can search for patterns across all of those sources, potentially accelerating target identification and narrowing the set of plausible therapies. The biological rationale is straightforward: if resistance is driven by a recurring molecular adaptation, then a system capable of learning that pattern across many patients may identify the next attack point earlier than a purely manual review.
Radiation, Surgery, and Systemic Therapy Can All Be Touched by the Same Data Layer
AI is not limited to diagnosis or discovery. The review describes a role in treatment planning for radiation and surgery, where precision depends on anatomy, tumor extent, and normal tissue constraints. In radiation oncology, AI may help contour targets, estimate organ-at-risk exposure, and standardize planning decisions across teams and centers. In surgery, algorithmic support may assist with preoperative risk estimation, procedural planning, and selection of patients most likely to benefit from intervention. These are not abstract conveniences. In thoracic oncology, a misjudged margin or an underestimated surgical risk can alter recurrence risk, perioperative complications, and long-term functional status.
Systemic therapy selection is another area where AI could assist clinicians. The review suggests that models may integrate pathology, imaging, molecular features, and clinical context to support individualized treatment choices. Lung cancer is already one of the most biomarker-driven cancers in routine practice, and the complexity is only increasing as more targeted agents and combination strategies enter use. Algorithms that synthesize heterogenous data may help identify patients more likely to respond to a given regimen or those at higher risk for toxicity. Even so, those outputs should be treated as decision support rather than decision replacement. The human team still has to weigh comorbidity burden, patient preference, symptom trajectory, and the subtleties that do not fit cleanly into a training set.
Continuous Monitoring Could Detect Resistance or Toxicity Earlier
A particularly important theme in the review is longitudinal monitoring. AI systems can be applied not only at baseline but also during treatment, where they may help detect emerging resistance or toxicity earlier than conventional review alone. That capacity is clinically attractive because the first sign of treatment failure is not always dramatic. It may appear as a tiny radiographic change, a slow biomarker rise, a subtle decline in performance status, or a toxicity pattern that becomes obvious only when viewed across several time points.
Continuous analysis could make those changes easier to see. If a model recognizes a pattern associated with progression or adverse effects, it may prompt closer surveillance, earlier intervention, or a treatment modification before the clinical picture worsens. In thoracic malignancies, where patients often receive layered care across oncology, pulmonology, pathology, radiology, surgery, and radiation oncology, this type of ongoing monitoring could reduce fragmentation. The challenge is to ensure that alerts improve care rather than overwhelm clinicians with noise.
Ethical and Regulatory Guardrails Will Decide Whether the Tools Scale Safely
The review gives substantial attention to governance, and for good reason. AI in oncology can only be trusted if it performs fairly across populations, settings, and disease subtypes. Bias mitigation is not a secondary issue; it is central to whether these tools can be deployed responsibly in real clinical practice. If the training data are narrow, the resulting model may reproduce historical inequities or fail in patients whose disease presentation differs from the populations used to build the algorithm.
Regulatory and clinical oversight also matter because thoracic oncology decisions often have high stakes and limited room for error. A false reassurance on imaging, a misleading pathology inference, or an overconfident treatment recommendation could have direct consequences. Human oversight remains essential, especially in decisions that involve complex tradeoffs or sensitive conversations with patients facing life-limiting disease. The review makes clear that AI should augment expertise rather than erode it. Clinical judgment, empathy, and accountability still belong to the care team.
The major unanswered issue is not whether AI can generate useful predictions in lung cancer. It already can in selected settings. The real test is whether those predictions can be integrated into everyday workflows without increasing inequity, confusion, or unsafe automation.