Artificial Intelligence-Driven Theranostics: Integrating Machine Learning For Precision Diagnosis And Personalized Therapy

Theranostics is a patient centered concept in which diagnostic characterization is directly linked to therapeutic selection and subsequent assessment of treatment response. The approach is particularly established in molecular imaging and nuclear medicine, where a diagnostic ligand can identify target expression and a related therapeutic ligand can deliver treatment to the same biological target. The central principle is not simply that diagnosis and therapy occur in the same pathway, but that information obtained at the diagnostic stage changes the therapeutic decision for an individual patient. This creates a natural setting for artificial intelligence because modern theranostic workflows generate high dimensional imaging, molecular, clinical, and longitudinal outcome data that are difficult to integrate using conventional rule based approaches [1]. AI encompasses computational methods capable of learning patterns from data, while ML refers to algorithms that learn relationships between input variables and outcomes. Deep learning extends this approach through multilayer neural networks that can learn hierarchical representations directly from complex data such as images, waveforms and text. In medicine, the value of these methods is increasingly demonstrated in image classification, segmentation, prognosis, clinical prediction, and treatment response modeling. Large scale studies have shown clinically relevant performance in retinal disease, skin cancer, breast cancer screening, pathology and electrocardiography [2]. These developments are important for theranostics because accurate target identification and quantitative phenotyping are prerequisites for individualized treatment. The growing availability of multimodal datasets creates an additional opportunity. A patient may simultaneously have CT, MRI, PET or SPECT images; pathology; laboratory measurements; genomic and transcriptomic profiles; medication history; treatment exposure; and longitudinal outcomes. AI can potentially combine these heterogeneous sources to estimate disease probability, identify biological subgroups, predict response, estimate toxicity and support treatment selection. Deep learning applied to electronic health records has demonstrated that large longitudinal records can be represented directly for prediction of multiple clinical events Similarly, radiomics has shown that quantitative imaging features can capture tumor characteristics beyond routine visual assessment [3]. The most useful way to conceptualize AI driven theranostics is as an integrated workflow rather than a single algorithm. The workflow begins with patient level data acquisition and quality control. Imaging data are reconstructed, normalized and segmented; molecular and clinical data are curated; and missingness, batch effects and site specific differences are addressed. AI models then generate quantitative features or predictions. These outputs can support diagnostic classification, target confirmation, risk stratification, treatment selection, dose estimation and response monitoring. New outcome data can subsequently be used to recalibrate or retrain models under controlled governance. This creates a feedback structure in which diagnostic information influences treatment and treatment outcomes improve future prediction [4].

The distinction between prediction and decision making is essential. An ML model may estimate the probability of treatment response, but a clinical decision also depends on contraindications, patient preference, toxicity risk, organ function, prior therapy, treatment availability and guideline recommendations. Therefore, an AI system should provide interpretable evidence, uncertainty estimates and relevant supporting variables rather than an unexplained command. In high consequence settings such as radiopharmaceutical therapy, individualized dosimetry and treatment planning must remain linked to physical and biological constraints [5]. The integration of AI into theranostics is also broader than oncology. The same principles can be applied to infectious diseases, cardiometabolic disorders, neurological disease and inflammatory disorders whenever diagnostic biomarkers can be connected to targeted intervention. However, oncology and nuclear medicine provide particularly clear examples because disease heterogeneity, molecular targets, imaging biomarkers and dose response relationships are central to treatment selection[6].

Figure 1. Integrated AI driven theranostic workflow

Summarizes major approaches and their theranostic applications. The important principle is that algorithm selection should follow the clinical question, data structure, sample size, interpretability requirement and validation strategy rather than being driven only by algorithmic novelty. Model development also requires careful separation of training, validation and testing data. Patient level splitting is essential because multiple images or visits from the same patient can otherwise leak information across datasets. External validation is more informative than an internal random split when the objective is clinical translation. Calibration should be assessed alongside discrimination because a model that ranks patients correctly can still produce misleading probabilities [18].

4. AI FOR PRECISION DIAGNOSIS AND QUANTITATIVE IMAGING

AI can support precision diagnosis at several levels: image reconstruction, lesion detection, lesion classification, quantitative measurement, differential diagnosis, and integrated reporting. In PET and SPECT, machine learning can be used to improve image quality from low count acquisitions, reduce noise, identify artifacts, and accelerate reconstruction. These applications may be particularly valuable when acquisition time, administered activity, or patient motion limits image quality. However, image enhancement algorithms should be validated against quantitative standards because visually attractive images can potentially alter measured uptake or obscure small lesions. Automated segmentation is one of the most important enabling technologies. Accurate delineation of tumors and organs allows calculation of metabolic tumor volume, total lesion activity, standardized uptake measures, organ activity, and spatially resolved dose. Segmentation algorithms can also reduce repetitive manual work and improve consistency between observers. Nevertheless, small lesions, low contrast lesions, heterogeneous uptake, postoperative anatomy, and lesions close to physiologically active organs remain challenging. Human review and correction remain important when segmentation affects treatment decisions [19]. AI supported quantitative imaging can also transform longitudinal assessment. Instead of comparing two scans only by visual impression, software can register serial images, match corresponding lesions, calculate changes in uptake and volume, and track lesion level trajectories. This enables a more granular view of treatment response, including mixed response in which some lesions regress while others remain stable or progress. Such information may be useful for adaptive therapy and for identifying patients who require a change in management. Medical imaging is one of the most established applications of AI. Deep learning has been used for classification, detection, segmentation, image reconstruction, denoising, registration and quantitative measurement. In diabetic retinopathy, deep learning achieved high diagnostic performance from retinal photographs . In retinal OCT, deep learning has been evaluated for diagnosis and referral . In breast imaging evaluation has demonstrated the potential of AI assisted screening Pathology provides another example. Whole slide images contain extremely large amounts of visual information, and weakly supervised learning can reduce the need for exhaustive pixel level annotation [20].

5. RADIOMICS AND RADIOGENOMICS

Radiomics converts medical images into large numbers of quantitative descriptors reflecting intensity, shape, texture, spatial relationships, and higher order patterns. Feature families can describe first order intensity distributions, shape characteristics, gray level cooccurrence relationships, runlength patterns, neighborhood dependence, and filtered or transformed image information. Machine learning can then identify combinations of features associated with diagnosis, molecular phenotype, treatment response, progression, or survival. Radiomics is highly sensitive to technical factors. Scanner manufacturer, acquisition protocol, reconstruction method, voxel dimensions, segmentation variability, preprocessing, and intensity discretization can change extracted features. Therefore, reproducibility requires standardized imaging protocols, feature quality assessment, test retest studies, harmonization, and independent validation. A feature that is statistically significant in one dataset should not automatically be interpreted as a stable biological biomarker [21]. Radiogenomics extends this concept by linking imaging phenotypes with genomic, transcriptomic, epigenomic, or other molecular information. The attraction is that imaging can provide a noninvasive and spatially distributed phenotype, whereas tissue sequencing provides molecular information that may be limited by sampling. Integrated models may help characterize tumor heterogeneity and infer biological processes that are difficult to observe from either imaging or genomics alone. Radiomics converts medical images into quantitative descriptors of intensity, shape, texture and spatial relationships. Aerts et al. demonstrated that radiomic features could characterize tumor phenotype and provide prognostic information. This approach is attractive for theranostics because it can provide information about tumor heterogeneity and disease phenotype without requiring additional invasive sampling. Radiogenomics links imaging phenotypes with genomic or molecular characteristics. It is especially relevant when a molecular biomarker is for treatment selection but tissue sampling is limited or incomplete [22].

6. AI FOR PERSONALIZED THERAPY AND DRUG RESPONSE PREDICTION

Personalized therapy requires estimation of both expected benefit and expected harm. AI models can combine demographic characteristics, disease stage, molecular markers, imaging phenotype, previous treatment exposure, laboratory variables, and patient specific factors to estimate response or toxicity. In drug development, machine learning can assist target identification, molecular property prediction, virtual screening, drug target interaction prediction, and optimization of candidate compounds. These approaches can shorten exploratory cycles, but computational predictions still require experimental and clinical confirmation. In theranostics, response prediction can be linked directly to the molecular target used for treatment. A model may estimate whether target positive disease is likely to respond, whether heterogeneous target expression may lead to incomplete treatment, or whether baseline tumor burden and organ reserve indicate a narrow therapeutic window. For radiopharmaceutical therapy, response prediction can potentially incorporate uptake intensity, lesion volume, spatial heterogeneity, absorbed dose, and prior cycle response rather than relying on a single imaging variable [23]. Adaptive treatment is a further extension of personalized therapy. After each treatment cycle, updated imaging and laboratory information can be used to revise the estimated response probability and toxicity risk. In principle, treatment intensity, cycle timing, or continuation can then be reconsidered using the patient's evolving data. Such adaptive approaches require prospective validation because retrospective associations do not demonstrate that changing therapy according to an algorithm improves outcomes. Personalized therapy aims to match an intervention with the biological and clinical characteristics of an individual patient. ML can integrate molecular alterations, imaging phenotype, previous treatment, laboratory values, pharmacogenomic variables and outcomes to estimate treatment response. Large pharmacogenomic studies have identified relationships between genomic features and drug sensitivity. Drug response prediction can be performed using patient derived molecular features or experimental cancer cell line datasets. Models may estimate sensitivity to individual agents, identify drug combinations or prioritize candidates for laboratory testing. Graph based approaches can represent molecular structures and drug target relationships . Protein structure prediction can also improve the computational representation of potential targets [24,25].

7. AI IN RADIOPHARMACEUTICAL THERAPY AND PERSONALIZED DOSIMETRY

Personalized dosimetry is one of the most clinically important areas for AI in radiotheranostics. The absorbed dose to a tumor or normal organ depends on activity distribution, biological clearance, physical decay, tissue mass, and time integrated activity. Traditional workflows may require several imaging time points and substantial manual processing. AI can assist registration, segmentation, activity quantification, time activity curve estimation, and dose map prediction, potentially reducing the time required for patient specific dosimetry. AI based dosimetry can operate at several spatial scales. Organ level approaches estimate average absorbed dose to organs, whereas lesion level and voxelbased methods preserve spatial heterogeneity. Voxel based approaches are particularly relevant when tumor dose is highly nonuniform because two lesions with the same mean dose may have different internal dose distributions. Deep learning models may estimate dose distributions from limited imaging time points, but their predictions should be compared with reference dosimetry and accompanied by uncertainty estimates [26]. Radiopharmaceutical therapy is a high value application of AI because the diagnostic and therapeutic stages can produce quantitative information about the same biological target. Personalized dosimetry estimates radiation absorbed by tumors and normal organs and may help optimize treatment intensity. Traditional dosimetry can require serial SPECT/CT imaging, image registration, segmentation, activity quantification, timeactivity curve fitting and computational dose calculation. These steps can be labor intensive. AI may reduce the burden through automated segmentation, image reconstruction, activity estimation and prediction of dose distributions [27,28].

8. AI FOR NANOTHERANOSTICS AND DRUG DISCOVERY

Nanotheranostics combines diagnostic and therapeutic functions within engineered nanoscale systems. Nanocarriers may be designed to transport drugs, imaging agents, radionuclides, nucleic acids, or combinations of these payloads. AI can support nanocarrier design by learning relationships between formulation variables and properties such as particle size, surface charge, encapsulation efficiency, stability, release kinetics, biodistribution, and cellular uptake. Machine learning can also help optimize formulation variables through design of experiments strategies and Bayesian optimization. Instead of evaluating every possible combination of materials and process parameters experimentally, models can prioritize formulations that are predicted to provide desirable properties. This can reduce experimental burden and identify nonlinear relationships that are difficult to detect using one factor at a time experimentation. Nevertheless, model predictions must remain linked to physicochemical characterization and biological testing [29,30]. For theranostic nanoparticles, AI may eventually integrate imaging, pharmacokinetics, molecular targeting, and treatment response to identify formulations suited to specific biological environments. Important translational barriers include manufacturing reproducibility, scale up, batch to batch consistency, long-term stability, toxicity assessment, regulatory characterization, and reproducible biodistribution. Computational optimization cannot substitute for these experimental and regulatory requirements [31,32]. Nanotheranostics combines diagnostic imaging and therapy within nanoscale systems. A nanoparticle may carry a therapeutic payload together with an imaging probe, enabling delivery and monitoring within a single platform. AI can assist formulation development by learning relationships among particle size, surface charge, drug loading, encapsulation efficiency, release kinetics and biological behavior. Drug discovery is another important area. Machine learning can predict drug target interactions, physicochemical properties and toxicity. Graph neural networks are suitable for molecular structures, while protein structure prediction can improve the representation of therapeutic targets. These approaches can prioritize compounds and reduce experimental search space [35].

9. AI FOR TREATMENT RESPONSE, TOXICITY AND LONGITUDINAL MONITORING

Treatment response assessment can be improved by combining quantitative imaging with temporal data. Instead of relying only on categorical response criteria, AI can learn continuous patterns involving lesion size, uptake, metabolic activity, heterogeneity, and dose response relationships. Such models may detect subtle changes before they become obvious on conventional visual assessment. However, the model endpoint must be clearly defined because early imaging changes can reflect inflammation, altered perfusion, or transient biological effects rather than durable tumor control [36]. Toxicity prediction is equally important because precision therapy must optimize the balance between tumor control and normal tissue safety. In radiopharmaceutical therapy, relevant variables may include kidney function, marrow reserve, administered activity, cumulative absorbed dose, prior therapies, and organ specific uptake. Machine learning can combine these variables to identify patients at increased risk of adverse effects. Prospective calibration and clinically meaningful thresholds are necessary before such models can guide treatment modification [37]. Longitudinal AI systems can maintain a patient specific trajectory rather than producing a single prediction. Each new imaging examination, laboratory result, treatment cycle, and adverse event can update the computational representation of the patient. This concept supports continuous monitoring and may eventually enable digital twin like simulations. Theranostics is inherently longitudinal because the treatment pathway generates new diagnostic information. AI can analyze serial imaging and laboratory measurements to detect changes in tumor volume, target uptake, heterogeneity and disease distribution [40].

10. CLINICAL TRANSLATION, ETHICS, SAFETY AND REGULATION

Clinical translation requires more than high accuracy on a retrospective test set. Dataset shift occurs when patient populations, scanners, acquisition protocols, disease prevalence or treatment practices change. External validation should therefore test meaningful variation in setting and population. Interpretability is important in high consequence decisions. Saliency maps, feature attribution and attention visualizations can provide clues about model behavior, but they do not automatically demonstrate biological validity. Explanations should be stable, clinically meaningful and capable of helping users recognize model errors [42,43]. Fairness must also be evaluated. An algorithm trained on unrepresentative data can produce unequal performance across patient groups. The well known analysis of a healthcare risk algorithm demonstrated that a seemingly useful prediction system can reproduce systematic disparities when the target outcome is an imperfect proxy for clinical need [45]. Theranostic AI should therefore evaluate performance across relevant populations and avoid inappropriate proxies. Privacy is especially important because theranostic datasets may combine imaging, genomic and longitudinal clinical information. Federated learning and privacy preserving computation may enable collaboration without centralizing raw data, although they introduce additional technical requirements. Governance should define intended use, contraindications, human oversight, audit trails, update procedures and post deployment monitoring. TRIPOD+AI provides updated reporting recommendations for clinical prediction models, while DECIDE AI, SPIRIT AI and CONSORT AI address early clinical evaluation and clinical trial reporting WHO guidance also emphasizes ethics, governance and regulatory considerations for AI in health [49,50].

11. RESEARCH GAPS AND FUTURE PERSPECTIVES

A major research gap is the limited availability of large, high quality, longitudinal theranostic datasets that link imaging, dosimetry, treatment exposure, toxicity, molecular data, and outcomes. Many current studies are retrospective and single center. Such designs are useful for hypothesis generation but can overestimate performance because of selection bias, spectrum bias, data leakage, and site specific technical characteristics. Multicenter datasets with standardized definitions and independent test cohorts are needed [55]. Another priority is uncertainty quantification. A clinically deployed system should ideally communicate when its prediction is outside the range represented in training data. Out of distribution detection, confidence estimation, conformal prediction, and task specific uncertainty modeling are promising approaches. In dosimetry, uncertainty may arise from segmentation, activity quantification, registration, calibration, pharmacokinetic assumptions, and physical dose calculation. Reporting only a single predicted dose can therefore provide an incomplete picture. A major gap is the limited number of prospective multicenter evaluations. Many studies remain retrospective and use technical accuracy as the primary endpoint. Future studies should determine whether AI changes clinically meaningful outcomes such as diagnostic accuracy, treatment selection, dosimetry quality, workflow time, toxicity, response or survival [58,60]. External and temporal validation should become routine. A model that performs well on an internal random split may fail when applied to another institution or later patient population. Calibration and uncertainty should be reported in addition to discrimination. Models should also detect out of distribution cases and defer uncertain cases to human review. Future systems should handle missing modalities without producing unjustified confidence. Standardized data models and interoperable pipelines will be necessary. Physics informed and biology informed AI may improve robustness [63,64]. In radiopharmaceutical therapy, models can incorporate radionuclide decay, activity distribution, anatomy and dose response relationships. In pharmacology, PK PD models can constrain adaptive dosing. Such hybrid systems may be safer and more interpretable than unconstrained black box models. These systems remain experimental and require prospective validation. Generative AI may also support radioligand design, image analysis, synthetic data generation and scientific information management, but hallucination and validation remain important concerns. Future clinical platforms should include automated data quality checks, uncertainty estimation, drift detection, version control, audit logs and predefined criteria for recalibration. The objective should be a human supervised closed loop in which AI supplies quantitative evidence while clinicians retain contextual responsibility for treatment decisions [66,67].

CONCLUSION

Artificial intelligence driven theranostics represents an emerging approach that combines advanced computational methods with precision diagnosis, targeted therapy, and individualized treatment monitoring. The integration of artificial intelligence and machine learning can transform complex clinical and biomedical data into clinically useful information for patient stratification, disease characterization, treatment selection, response prediction, and toxicity assessment. In medical imaging, deep learning can support lesion detection, segmentation, quantitative image analysis, radiomics, and longitudinal assessment, while radiogenomics can connect imaging phenotypes with molecular characteristics. These capabilities may improve understanding of tumor heterogeneity and support more individualized therapeutic strategies. Radiopharmaceutical theranostics provides an important application in which artificial intelligence can assist target identification, quantitative imaging, personalized dosimetry, treatment planning, response assessment, and adaptive therapy. AI assisted dosimetry may reduce computational workload and potentially facilitate patient specific treatment optimization. Similarly, machine learning can contribute to nanotheranostic formulation, drug discovery, drug response prediction, and identification of clinically relevant biomarkers. Despite these opportunities, important challenges remain, including limited multicenter datasets, data heterogeneity, algorithmic bias, lack of external validation, interpretability, uncertainty estimation, regulatory requirements, data privacy, and integration into clinical workflows. High retrospective accuracy alone cannot establish clinical effectiveness. Prospective studies, standardized reporting, independent validation, multidisciplinary collaboration, and continuous monitoring are required for responsible implementation. Overall, AI driven theranostics has the potential to establish a more quantitative, adaptive, and patient centered model of precision medicine. Future research should focus on robust multimodal models, personalized dosimetry, explainable and trustworthy AI, privacy preserving collaboration, and prospective clinical evaluation. Successful translation will depend on demonstrating that AI supported decisions provide measurable improvements in treatment personalization, safety, efficiency, and patient outcomes.

REFERENCES

1. Brosch Lenz J, Yousefirizi F, Zukotynski K, Beauregard JM, Gaudet V, Saboury B, Rahmim A, Uribe C. Role of artificial intelligence in theranostics: toward routine personalized radiopharmaceutical therapies. PET clinics. 2021 Oct 1;16(4):627-41.
2. Bilgin GB, Bilgin C, Burkett BJ, Orme JJ, Childs DS, Thorpe MP, Halfdanarson TR, Johnson GB, Kendi AT, Sartor O. Theranostics and artificial intelligence: new frontiers in personalized medicine. Theranostics. 2024 Mar 25;14(6):2367.
3. Currie GM, Rohren E. The role of artificial intelligence in theranostics. Journal of Nuclear Medicine Technology. 2025 Dec 1;53(Supplement 1):82S-91S.
4. Naik K, Goyal RK, Foschini L, Chak CW, Thielscher C, Zhu H, Lu J, Lehár J, Pacanoswki MA, Terranova N, Mehta N. Current status and future directions: the application of artificial intelligence/machine learning for precision medicine. Clinical Pharmacology & Therapeutics. 2024 Apr;115(4):673-86.
5. Topol EJ. High performance medicine: the convergence of human and artificial intelligence. Nature medicine. 2019 Jan;25(1):44-56.
6. Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, Liu PJ, Liu X, Marcus J, Sun M, Sundberg P. Scalable and accurate deep learning with electronic health records. NPJ digital medicine. 2018 May 8;1(1):18.
7. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S. Dermatologist level classification of skin cancer with deep neural networks. nature. 2017 Feb 2;542(7639):115-8.
8. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. jama. 2016 Dec 13;316(22):2402-10.
9. De Fauw J, Ledsam JR, Romera Paredes B, Nikolov S, Tomasev N, Blackwell S, Askham H, Glorot X, O’Donoghue B, Visentin D, Van Den Driessche G. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature medicine. 2018 Sep;24(9):1342-50.
10. Kermany DS, Goldbaum M, Cai W, Valentim CC, Liang H, Baxter SL, McKeown A, Yang G, Wu X, Yan F, Dong J. Identifying medical diagnoses and treatable diseases by image-based deep learning. cell. 2018 Feb 22;172(5):1122-31.
11. McKinney SM, Sieniek M, Godbole V, Godwin J, Antropova N, Ashrafian H, Back T, Chesus M, Corrado GS, Darzi A, Etemadi M. International evaluation of an AI system for breast cancer screening. Nature. 2020 Jan 2;577(7788):89-94.
12. Campanella G, Hanna MG, Geneslaw L, Miraflor A, Werneck Krauss Silva V, Busam K J, Brogi E, Reuter VE, Klimstra DS, Fuchs TJ. Clinical grade computational pathology using weakly supervised deep learning on whole slide images. Nature medicine. 2019 Aug;25(8):1301-9.
13. Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, Ng AY. Cardiologist level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nature medicine. 2019 Jan;25(1):65-9.
14. Attia ZI, Noseworthy PA, Lopez Jimenez F, Asirvatham SJ, Deshmukh AJ, Gersh BJ, Carter RE, Yao X, Rabinstein AA, Erickson BJ, Kapa S. An artificial intelligence enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction. The Lancet. 2019 Sep 7;394(10201):861-7.
15. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, Bussink J, Monshouwer R, Haibe Kains B, Rietveld D, Hoebers F. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nature communications. 2014 Jun 3;5(1):4006.
16. Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW, Greninger P, Thompson IR, Luo X, Soares J, Liu Q. Systematic identification of genomic markers of drug sensitivity in cancer cells. Nature. 2012 Mar 29;483(7391):570-5.
17. Menden MP, Iorio F, Garnett M, et al. Machine learning prediction of cancer cell sensitivity to drugs based on genomic and chemical properties. PLoS One. 2013;8:e61318.
18. Iorio F, Knijnenburg TA, Vis DJ, Bignell GR, Menden MP, Schubert M, Aben N, Gonçalves E, Barthorpe S, Lightfoot H, Cokelaer T. A landscape of pharmacogenomic interactions in cancer. Cell. 2016 Jul 28;166(3):740-54.
19. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, Bridgland A. Highly accurate protein structure prediction with AlphaFold. nature. 2021 Aug 26;596(7873):583-9.
20. Tunyasuvunakool K, Adler J, Wu Z, Green T, Zielinski M, Žídek A, Bridgland A, Cowie A, Meyer C, Laydon A, Velankar S. Highly accurate protein structure prediction for the human proteome. Nature. 2021 Aug 26;596(7873):590-6.
21. Li Y, Liang W, Peng L, Zhang D, Yang C, Li KC. Predicting drug target interactions via dual stream graph neural network. IEEE/ACM Transactions on Computational Biology and Bioinformatics. 2022 Sep 8;21(4):948-58.
22. Strosberg J, El Haddad G, Wolin E, Hendifar A, Yao J, Chasen B, Mittra E, Kunz PL, Kulke MH, Jacene H, Bushnell D. Phase 3 trial of 177Lu Dotatate for midgut neuroendocrine tumors. New England Journal of Medicine. 2017 Jan 12;376(2):125-35.
23. Strosberg JR, Caplin ME, Kunz PL, Ruszniewski PB, Bodei L, Hendifar A, Mittra E, Wolin EM, Yao JC, Pavel ME, Grande E. NETTER 1 investigators. 177 Lu Dotatate plus long-acting octreotide versus high-dose long-acting octreotide in patients with midgut neuroendocrine tumours (NETTER-1): final overall survival and long term safety results from an open label, randomised, controlled, phase 3 trial. Lancet Oncol. 2021 Dec;22(12):1752-63.
24. Sartor O, De Bono J, Chi KN, Fizazi K, Herrmann K, Rahbar K, Tagawa ST, Nordquist LT, Vaishampayan N, El Haddad G, Park CH. Lutetium 177 PSMA 617 for metastatic castration resistant prostate cancer. New England Journal of Medicine. 2021 Sep 16;385(12):1091-103.
25. Hofman MS, Emmett L, Sandhu S, Iravani A, Joshua AM, Goh JC, Pattison DA, Tan TH, Kirkwood ID, Ng S, Francis RJ. [177Lu] Lu-PSMA-617 versus cabazitaxel in patients with metastatic castration resistant prostate cancer (TheraP): a randomised, open label, phase 2 trial. The Lancet. 2021 Feb 27;397(10276):797-804.
26. Buteau JP, Martin AJ, Emmett L, Iravani A, Sandhu S, Joshua AM, Francis RJ, Zhang AY, Scott AM, Lee ST, Azad AA. PSMA and FDG-PET as predictive and prognostic biomarkers in patients given [177Lu] Lu PSMA 617 versus cabazitaxel for metastatic castration resistant prostate cancer (TheraP): a biomarker analysis from a randomised, open label, phase 2 trial. The Lancet Oncology. 2022 Nov 1;23(11):1389-97.
27. Sandström M, Garske U, Granberg D, Sundin A, Lundqvist H. Individualized dosimetry in patients undergoing therapy with 177Lu DOTA D Phe1 Tyr3 octreotate. European journal of nuclear medicine and molecular imaging. 2010 Feb;37(2):212-25.
28. Mansouri Z, Salimi Y, Akhavanallaf A, Shiri I, Teixeira EP, Hou X, Beauregard JM, Rahmim A, Zaidi H. Deep transformer based personalized dosimetry from SPECT/CT images: a hybrid approach for [177Lu] Lu DOTATATE radiopharmaceutical therapy. European journal of nuclear medicine and molecular imaging. 2024 May;51(6):1516-29.
29. Bednarz B. Theranostics and patient specific dosimetry. In Seminars in radiation oncology 2023 Jul 1 (Vol. 33, No. 3, pp. 317-326). WB Saunders.
30. Woo SK. The role of artificial intelligence in advancing theranostics dosimetry for cancer therapy: a review. Nuclear Medicine and Molecular Imaging. 2025 Oct;59(5):329-41.
31. Montes AL, Yousefirizi F, Chen Y, Salimi Y, Seifert R, Afshar Oromieh A, Uribe C, Rominger A, Zaidi H, Rahmim A, Shi K. Artificial intelligence for simplified patient centered dosimetry in radiopharmaceutical therapies. PET clinics. 2026 Jan 1;21(1):73-88.
32. Pinker K, Chin J, Melsaether AN, Morris EA, Moy L. Precision medicine and radiogenomics in breast cancer: new approaches toward diagnosis and treatment. Radiology. 2018 Jun;287(3):732-47.
33. Saxena S, Jena B, Gupta N, Das S, Sarmah D, Bhattacharya P, Nath T, Paul S, Fouda MM, Kalra M, Saba L. Role of artificial intelligence in radiogenomics for cancers in the era of precision medicine. Cancers. 2022 Jun 9;14(12):2860.
34. Trivizakis E, Papadakis GZ, Souglakos I, Papanikolaou N, Koumakis L, Spandidos DA, Tsatsakis A, Karantanas AH, Marias K. Artificial intelligence radiogenomics for advancing precision and effectiveness in oncologic care. International journal of oncology. 2020 May 11;57(1):43-53.
35. He W, Huang W, Zhang L, Wu X, Zhang S, Zhang B. Radiogenomics: bridging the gap between imaging and genomics for precision oncology. MedComm. 2024 Sep;5(9):e722.
36. Ros H, Chan N, Cook MT, Shorthouse D. Artificial intelligence and machine learning guided optimization in drug delivery. Advanced Drug Delivery Reviews. 2026 Jan 22:115781.
37. Shen C, Zhang M, Lu M, Chang E, Gao Z, Ban W, Liu Q, Zuo Z, Jiang C. Machine learning empowered formulation design, optimization and characterization of nanoparticulate drug delivery systems: Current applications, challenges, and future perspectives. Acta Pharmaceutica Sinica B. 2025 Dec 10.
38. Lucas I, Sousa J, Vitorino C. Integrating artificial intelligence into drug delivery systems: Formulation development and current challenges. European Journal of Pharmaceutics and Biopharmaceutics. 2026 Jun 3:115133.
39. Alsaqabi AA, Almoutairi AA, Alnehari F, Alanazi AN, Aldugiem R, Alsaeed Y, Alotaibi S. Artificial Intelligence Driven Nanomedicine: From Drug Formulation and Nanocarrier Design to Clinical Translation. Pharmaceutics. 2026 Jul 11;18(7):845.
40. Collins GS, Moons KG, Dhiman P, Riley RD, Beam AL, Van Calster B, Ghassemi M, Liu X, Reitsma JB, Van Smeden M, Boulesteix AL. TRIPOD+ AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. bmj. 2024 Apr 16;385.
41. Cohen JF, Bossuyt PM. TRIPOD+ AI: an updated reporting guideline for clinical prediction models. bmj. 2024 Apr 16;385.
42. Lekadir K, Frangi AF, Porras AR, Glocker B, Cintas C, Langlotz CP, Weicken E, Asselbergs FW, Prior F, Collins GS, Kaissis G. FUTURE AI: international consensus guideline for trustworthy and deployable artificial intelligence in healthcare. bmj. 2025 Feb 5;388.
43. Kelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D. Key challenges for delivering clinical impact with artificial intelligence. BMC medicine. 2019 Oct 29;17(1):195.
44. Haibe Kains B, Adam GA, Hosny A, Khodakarami F, Massive Analysis Quality Control (MAQC) Society Board of Directors Shraddha Thakkar 35 Kusko Rebecca 36 Sansone Susanna Assunta 37 Tong Weida 35 Wolfinger Russ D. 38 Mason Christopher E. 39 Jones Wendell 40 Dopazo Joaquin 41 Furlanello Cesare 42, Waldron L, Wang B, McIntosh C, Goldenberg A, Kundaje A, Greene CS. Transparency and reproducibility in artificial intelligence. Nature. 2020 Oct 15;586(7829):E14-6.
45. Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019 Oct 25;366(6464):447-53.
46. Beam AL, Kohane IS. Big data and machine learning in health care. Jama. 2018 Apr 3;319(13):1317-8.
47. Miotto R, Wang F, Wang S, Jiang X, Dudley JT. Deep learning for healthcare: review, opportunities and challenges. Briefings in bioinformatics. 2018 Nov;19(6):1236-46.
48. Rajpurkar P, Chen E, Banerjee O, Topol EJ. AI in health and medicine. Nature medicine. 2022 Jan;28(1):31-8.
49. Litjens G, Kooi T, Bejnordi BE, Setio AA, Ciompi F, Ghafoorian M, Van Der Laak JA, Van Ginneken B, Sánchez CI. A survey on deep learning in medical image analysis. Medical image analysis. 2017 Dec 1;42:60-88.
50. LeCun Y, Bengio Y, Hinton G. Deep learning. nature. 2015 May 28;521(7553):436-44.
51. Gage BF, Eby C, Johnson JA, Deych E, Rieder MJ, Ridker PM, Milligan PE, Grice G, Lenzini P, Rettie AE, Aquilante CL. Use of pharmacogenetic and clinical factors to predict the therapeutic dose of warfarin. Clinical Pharmacology & Therapeutics. 2008 Sep;84(3):326-31.
52. Millican EA, Lenzini PA, Milligan PE, Grosso L, Eby C, Deych E, Grice G, Clohisy JC, Barrack RL, Burnett RS, Voora D. Genetic based dosing in orthopedic patients beginning warfarin therapy. Blood, The Journal of the American Society of Hematology. 2007 Sep 1;110(5):1511-5.
53. Kimmel SE, French B, Kasner SE, Johnson JA, Anderson JL, Gage BF, Rosenberg YD, Eby CS, Madigan RA, McBane RB, Abdel Rahman SZ. A pharmacogenetic versus a clinical algorithm for warfarin dosing. New England Journal of Medicine. 2013 Dec 12;369(24):2283-93.
54. Vasey B, Nagendran M, Campbell B, Clifton DA, Collins GS, Denaxas S, Denniston AK, Faes L, Geerts B, Ibrahim M, Liu X. Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE AI. Nature medicine. 2022 May;28(5):924-33.
55. Liu X, Chan AW, Denniston AK, Calvert MJ. Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT AI extension Diretrizes para protocolos de ensaios clínicos com intervenções que utilizam inteligência artificial: a extensão SPIRIT AI. Revista Panamericana de Salud Publica Pan American Journal of Public Health. 2023 Feb 1;48:e12-.
56. Ibrahim H, Liu X, Rivera SC, Moher D, Chan AW, Sydes MR, Calvert MJ, Denniston AK. Reporting guidelines for clinical trials of artificial intelligence interventions: the SPIRIT AI and CONSORT-AI guidelines. Trials. 2021 Jan 6;22(1):11.
57. Guidance WH. Ethics and governance of artificial intelligence for health. World Health Organization. 2021 Jun 28;1.
58. World Health Organization. Regulatory considerations on artificial intelligence for health. World Health Organization; 2023 Oct 19.
59. Topol E. The topol review. Preparing the healthcare workforce to deliver the digital future. 2019 Feb:1-48.
60. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. Journal of the American College of Cardiology. 2017 May 30;69(21):2657-64.
61. Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, Ng AY. Cardiologist level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nature medicine. 2019 Jan;25(1):65-9.
62. Hwang EJ, Nam JG, Lim WH, Park SJ, Jeong YS, Kang JH, Hong EK, Kim TM, Goo JM, Park S, Kim KH. Deep learning for chest radiograph diagnosis in the emergency department. Radiology. 2019 Dec;293(3):573-80.
63. Campanella G, Hanna MG, Geneslaw L, Miraflor A, Werneck Krauss Silva V, Busam KJ, Brogi E, Reuter VE, Klimstra DS, Fuchs TJ. Clinical grade computational pathology using weakly supervised deep learning on whole slide images. Nature medicine. 2019 Aug;25(8):1301-9.
64. Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, Cui C, Corrado G, Thrun S, Dean J. A guide to deep learning in healthcare. Nature medicine. 2019 Jan;25(1):24-9.
65. Liu X, Faes L, Kale AU, Wagner SK, Fu DJ, Bruynseels A, Mahendiran T, Moraes G, Shamdas M, Kern C, Ledsam JR. A comparison of deep learning performance against health care professionals in detecting diseases from medical imaging: a systematic review and meta analysis. The lancet digital health. 2019 Oct 1;1(6):e271-97.
66. Banala M, Ezuddin S, Foley M, Kuker R. Artificial intelligence across oncologic theranostics: evidence for patient stratification, dosimetry, and adaptive radiopharmaceutical therapy. Frontiers in Nuclear Medicine. 2026 Aug 27;6:1935247.
67. Wahl RL, Dewaraja YK. Dosimetry driven personalization of radiopharmaceutical therapies: evidence and opportunities. Journal of Nuclear Medicine. 2026 Sep 1;67(Supplement 2):60S-78S