Click here to

Session: Therapy General ePoster Viewing [Return to Session]

A Machine Learning Prognostic Model for the Assessment of 177-Lu-DOTATATE Therapy Using Dose and Radiomics Features Analysis

D Plachouris1, V Eleftheriadis2, T Nanos1, P Papadimitroulas2, L Vergnaud3, N Papathanasiou4, D Sarrut3, G C Kagadis1*, (1) Department of Medical Physics, School of Medicine, University of Patras, Rion, GR,(2) Bioemission Technology Solutions - BIOEMTECH, Athens, GR, (3) CREATIS, CNRS, Universite de Lyon, Lyon, FR, (4) Department of Nuclear Medicine, School of Medicine, University of Patras, Rion, GR


PO-GePV-T-15 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: A standardized dosimetry procedure is the cornerstone in individualized radioimmunotherapy which, could bring improvements in the therapeutic outcome. Prediction of the absorbed dose to every tissue is a key element to define organ-at-risk (OAR) absorbed dose safety limits.

Methods: Data from 13 patients treated for NETs with 177-Lu DOTATATE were collected and 1,705 radiomics features were extracted combined with dose values calculated through direct MC calculations. A regression prognostic model was designed based on ML algorithms for predicting the absorbed dose in Liver and Spleen ROIs before, during, and after the treatment, and thus consequently predict any possible radiotoxicity. The radiomics’ features investigated in the present study were extracted, with the use of PyRadiomics library, for all possible multimodal images available in the clinical practice of 177-Lu DOTATATE treatment planning (PET, SPECT, and CT scans).

Results: Several ML regressions models were evaluated. The best predictive model achieved an absolute average dose accuracy of 62.50% between pre-therapy 68Ga and every post-therapy 177Lu cycle treatment and was based on the GradientBoost Regressor model. Our model achieved an average of 64.23% and 60.77% dose predictive accuracy for Spleen and Liver respectively. The average absorbed dose differences between MC simulation and our prediction model showed: Ga-Cycle1: 20.91%, Ga-Cycle2: -18.00%, Ga-Cycle3: 1.23%, Ga-Cycle4: -13.83%. Cycle1-Cycle2: 30.15%, Cycle1-Cycle3: -35.06%, Cycle1-Cycle4: 2.45%. Cycle2-Cycle3: 3.53%, Cycle2-Cycle4: -34.03%. Cycle3-Cycle4: -15.93%.

Conclusion: The combination of radiomics features and dose values shows potential utility for prognosis and therapy response evaluation. In modern personalized medicine, more ‘radiodosiomic’ studies are needed to reveal predictive radiomics signatures and dose expression patterns for broadening the potential of personalized treatment planning into nuclear medicine. These features can provide recurrence-related information and could be helpful in clinical decision-making, especially regarding dose escalation.

Funding Support, Disclosures, and Conflict of Interest: This research was financed by the European Regional Development Fund(ERDF), Greek General Secretariat for Research and Innovation, Operational Programme 'Competitiveness, Entrepreneurship and Innovation' (EPAnEK), under the frame of ERA PerMed (project POPEYE T11EPA4-00055).


Monte Carlo, Radioimmunotherapy


IM/TH- Radiopharmaceutical Therapy: Dose estimation: Monte Carlo

Contact Email