How Artificial Intelligence Can Redefine Rehabilitation for Cancer Survivors
By Hari Vennelakanti, Oncology Physical Therapist | Health Informatics MS Candidate, UT Southwestern Medical Center
The Personalized Challenge in Oncology Rehabilitation
The physical recovery journey for cancer survivors, commonly referred to as oncology rehabilitation, has long been constrained by a fundamental paradox: clinicians are attempting to apply standardized, one-size-fits-all treatment protocols to what is fundamentally a highly individualized and heterogeneous disease experience. Each cancer survivor emerges from treatment with a unique constellation of challenges that reflect not only their specific cancer type and treatment regimen but also their baseline health status, genetic predispositions, and psychosocial circumstances.
These survivors frequently grapple with a complex array of debilitating symptoms including persistent fatigue that can endure for months or years post-treatment, lymphedema characterized by chronic swelling resulting from lymph node removal or radiation damage, sarcopenia manifested as progressive muscle mass loss and functional decline, cardiorespiratory compromise secondary to cardiotoxic chemotherapy agents or chest radiation, peripheral neuropathy causing pain and functional limitations, and cognitive impairment colloquially termed ‘chemo brain.’ The severity and combination of these sequelae vary dramatically from patient to patient, and even within a single patient, symptoms fluctuate substantially based on factors such as activity level, environmental stressors, and recovery phase. Despite this obvious heterogeneity and dynamic variability, the prescribed rehabilitation care often remains frustratingly static, delivered according to rigid protocols that fail to adapt to the patient’s evolving needs and capabilities.
Artificial Intelligence (AI) represents the transformative solution to overcoming this longstanding clinical challenge. By integrating and synthesizing previously fragmented data streams, including electronic health records, wearable device metrics, patient-reported outcomes, imaging studies, and laboratory values, AI systems can generate a comprehensive, real-time clinical picture of each individual patient. This holistic data integration enables the delivery of just-in-time clinical decision support that adapts dynamically to the patient’s current status. Furthermore, AI algorithms can provide critical automated alerts to healthcare providers when concerning patterns emerge, such as flagging a precipitous and uncharacteristic decline in a patient’s daily activity level that might herald an impending systemic infection, metabolic decompensation, or the need for urgent medication adjustment before clinical deterioration necessitates emergency intervention or hospitalization.
Today, we are witnessing a powerful and accelerating paradigm shift in cancer care. Artificial Intelligence (AI) has the potential to transform oncology rehabilitation from a traditionally reactive, generalized, and protocol-driven set of exercises into a sophisticated, highly personalized, predictive, and readily accessible system of comprehensive care. It is crucial to emphasize that this transformation is not about robots or algorithms replacing human therapists and clinicians. Rather, it is fundamentally about augmenting and extending the capabilities of the human clinician by providing them with what can be conceptualized as an intelligent system that processes vast amounts of data, identifies subtle patterns, and generates actionable insights, thereby ensuring that the prescribed treatment precisely meets each patient exactly where they are on any given day of their recovery journey.
AI for Early Detection and Diagnosis of Post-Cancer Complications
One of the most clinically impactful and promising roles that AI is currently fulfilling in oncology rehabilitation is addressing the persistent problem of early detection for critical post-cancer morbidities that have historically evaded timely identification. Traditional diagnostic methods for debilitating conditions such as lymphedema, the chronic and often progressive swelling that can develop following lymph node removal or radiation therapy, have relied heavily on tools such as circumferential tape measurements or bioimpedance analysis devices. Unfortunately, these conventional approaches often only succeed in confirming the diagnosis once the pathological condition is already firmly established and has progressed to a clinical stage.
Artificial Intelligence is demonstrating remarkable capability in identifying these conditions during their subclinical stages, the critical window before objective physical changes become readily apparent to conventional measurement techniques. A groundbreaking study conducted by Fu and colleagues (2018) utilized an innovative mobile health platform to systematically collect Patient-Reported Outcomes, which capture patients’ subjective symptom experiences, from a cohort of breast cancer survivors. By inputting these nuanced symptom clusters, including sensations of heaviness, tightness, aching, and altered sensation into a sophisticated Neural Network algorithm, the research team achieved a diagnostic accuracy approaching 94 percent for lymphedema detection. This performance represents a significant and clinically meaningful improvement over standard bioimpedance methodologies, convincingly demonstrating AI’s ability to recognize complex, multidimensional patterns in how patients describe what they are experiencing, and to do so considerably earlier in the disease process before volumetric changes become measurable through traditional means.
Similarly, Lim and Song (2025) transformative applications of AI are emerging in the diagnosis and screening of sarcopenia, which is characterized by the progressive loss of skeletal muscle mass and functional capacity. Traditionally, accurate assessment of muscle mass has required expensive, radiation-exposing imaging modalities such as computed tomography scans or dual-energy X-ray absorptiometry. In a stunning technical breakthrough, researchers were able to test breast cancer survivors for sarcopenia with an astounding 95 percent accuracy by combining a basic consumer-grade gaming camera with a specialized machine learning model called XGBoost. The critical insight underlying this success was the AI system’s capacity for objective, high-precision kinematic tracking during functional movement tasks. The algorithm discovered that sarcopenic patients, even in subclinical stages, demonstrate subtle compensatory movement patterns. Specifically, they unconsciously limit their Right Knee Flexion range of motion during exercise activities to compensate for underlying quadriceps muscle weakness. This subtle movement avoidance strategy serves as an early biomechanical signal of systemic muscle failure that the unaided human eye frequently overlooks, thereby enabling earlier identification and intervention before substantial functional decline occurs.
The future of oncology rehabilitation is definitely not one of complete automation.
Driving Functional Recovery, Treatment Adherence, and Comprehensive Support
Beyond its diagnostic capabilities, AI is increasingly playing a central and indispensable role in the prescription, monitoring, and optimization of therapeutic exercise interventions. The ultimate goal of oncology rehabilitation extends far beyond mere survival, it encompasses the restoration of meaningful quality of life, functional independence, and cardiorespiratory capacity that enables patients to return to valued activities and social roles.
The critical importance of AI-driven interventions in achieving functional gains was compellingly demonstrated in a rigorous 2025 Randomized Controlled Trial conducted by Li and colleagues, which specifically examined Digital Therapeutics applications in the emerging field of cardio-oncology. In this study, lung cancer survivors who participated in a sophisticated home-based AI program that incorporated wearable devices for continuous real-time heart rate monitoring and automated exercise prescription adjustment experienced a statistically significant and clinically meaningful increase in their peak oxygen consumption a gold-standard measure of cardiorespiratory fitness of 3.66 milliliters per kilogram per minute when compared to the minimal improvement observed in the standard-care control group.
Perhaps the most compelling and unexpected finding from this trial was the extraordinary level of treatment adherence achieved in the AI intervention group. Working with a patient population that is notoriously challenged by treatment-related
fatigue, geographic barriers to clinic access, transportation difficulties, and competing demands, the researchers observed compliance rates that exceeded 100 percent. This remarkable statistic indicates that patients were not merely completing their prescribed exercise regimens but were so engaged with the intelligent, adaptive program that they voluntarily performed more exercise sessions than originally assigned. Most importantly, this high-intensity engagement was achieved with zero exercise-related adverse events, conclusively demonstrating that a thoughtfully designed, AI-driven therapeutic program can simultaneously be safe, clinically effective, and highly engaging for this vulnerable patient population.
The journey of recovery from cancer necessarily encompasses the whole patient as a complex biopsychosocial being, extending far beyond purely physical rehabilitation of muscles and cardiovascular capacity to include critical domains such as nutritional optimization, mental health support, and the effective management of anxiety and psychological distress that frequently intensifies during evening hours and weekends when professional clinical support is typically unavailable. AI-powered virtual health assistants and conversational chatbots are successfully bridging this critical support gap by offering around-the-clock availability that is particularly invaluable for patients residing in rural, remote, or medically underserved geographic areas. These intelligent systems efficiently handle the substantial volume of routine informational questions, deliver automated medication reminders, facilitate real-time symptom tracking and reporting, and provide immediate psychosocial support, thereby liberating human clinicians to concentrate their expertise and emotional energy on complex clinical decision-making and the provision of empathetic, personalized care. Furthermore, dedicated AI-enabled mobile applications are now being deployed to function as personalized nutritionists and mental health coaches, generating individualized dietary plans and wellness strategies specifically tailored to the unique metabolic, nutritional, and psychological needs of cancer survivors, rather than offering generic advice that fails to account for their specific medical context.
Overcoming Data Integration Challenges and Addressing Ethical Imperatives
Despite the impressive feasibility of wearable technology platforms, with adherence rates consistently exceeding 80 percent across multiple research studies, the current landscape is significantly compromised by a critical limitation that has been termed the Offline Problem. A comprehensive systematic review by Chow and colleagues published in 2024 revealed the sobering finding that 97 percent of wearable device studies utilized the collected physiological and activity data retrospectively, analyzing it only after the study period had concluded. To unlock the full transformative potential of wearable technologies and AI-driven monitoring systems, the field must urgently transition from this offline, retrospective data collection paradigm to an online, real-time
An integration model that enables AI algorithms to identify concerning patterns and flag potential complications as they develop. Under the current retrospective approach, data indicating a patient’s deteriorating heart rate variability or progressive decline in daily movement activity is only analyzed and flagged days or even weeks after the patient has already experienced clinical deterioration requiring emergency department evaluation or hospital admission, representing a fundamental failure to realize the preventive promise of these technologies.
This technological transformation, however promising, is not without significant ethical challenges that must be thoughtfully addressed. A particularly critical concern revolves around the potential misuse of health data gathered from wearable monitoring devices. Specifically, there exists a legitimate risk that detailed information regarding patient activity patterns, exercise compliance, and symptom reporting could be utilized in punitive ways to determine and penalize patients for perceived non-compliance with prescribed treatment and rehabilitation plans. This troubling possibility raises profound questions about fundamental patient rights, including privacy, autonomy, informed consent, and the potential for discriminatory practices by insurance providers or healthcare systems that might deny coverage or increase premiums based on compliance metrics derived from AI analysis of wearable device data. The establishment of robust data governance frameworks, transparent consent mechanisms that clearly explain data usage, and comprehensive legal safeguards are absolutely essential to effectively mitigate these risks and ensure that AI remains a tool for patient empowerment and improved health outcomes rather than a mechanism for coercion or discrimination.
The Human-in-the-Loop Future of Oncology Rehabilitation
The future of oncology rehabilitation is definitely not one of complete automation. Rather, the optimal and most ethically sound paradigm is what has been termed the human-in-the-loop model. While AI demonstrates exceptional capabilities in objectifying complex movement patterns, continuously tracking multidimensional symptom profiles, and analyzing vast datasets to identify subtle patterns that would overwhelm human cognitive capacity, it simultaneously carries inherent risks and limitations. These include persistent challenges related to data privacy and security, the documented phenomenon of algorithmic hallucination whereby AI systems can confidently generate incorrect recommendations that could prove harmful or even life-threatening if implemented without human oversight, and perhaps most fundamentally, what researchers have termed the empathy gap the inherent inability of artificial systems to authentically replicate human compassion, emotional understanding, and the capacity to fully comprehend the profound existential and psychological weight of a cancer diagnosis.
The appropriate role of AI is therefore to serve as a powerful augmentation and extension of the clinician’s inherent capabilities rather than as a replacement for human judgment and care. By delegating to AI the computationally intensive tasks of data processing, pattern recognition, and objective biomechanical analysis, the oncology rehabilitation specialist is liberated to focus their professional expertise, clinical experience, and emotional energy on what matters most: ensuring patient safety through vigilant oversight, exercising nuanced clinical judgment in complex situations where guidelines are insufficient, and providing the essential human connection, empathetic presence, and personalized encouragement that remain irreplaceable components of delivering high-quality, patient-centered care. This collaborative human-AI model holds particular promise for extending specialized oncology rehabilitation care through telehealth and telerehabilitation service delivery platforms to rural and medically underserved patient populations, thereby making previously inaccessible specialized care more equitable, geographically distributed, and available to all cancer survivors regardless of their location or socioeconomic circumstances.
References
Chow, R., Drkulec, H., Im, J. H. B., Tsai, J., Nafees, A., Kumar, S., Hou, T., Fazelzad, R., Leighl, N. B., Krzyzanowska, M., Wong, P., & Raman, S. (2024). The Use of Wearable Devices in Oncology Patients: A Systematic Review. The oncologist, 29(4), e419–e430. https://doi.org/10.1093/oncolo/oyad305
Fu, M. R., Wang, Y., Li, C., Qiu, Z., Axelrod, D., Guth, A. A., Scagliola, J., Conley, Y., Aouizerat, B. E., Qiu, J. M., Yu, G., Van Cleave, J. H., Haber, J., & Cheung, Y. K. (2018). Machine learning for detection of lymphedema among breast cancer survivors. mHealth, 4, 17. https://doi.org/10.21037/mhealth.2018.04.02
Lim, B., & Song, W. (2025). Development and validation of machine learning models for classifying cancer-related sarcopenia using Kinect-based mixed-reality exercises in breast cancer survivors. Translational cancer research, 14(7), 4208–4218. https://doi.org/10.21037/tcr-2024-2337