Aims: i) To non-invasively stratify pancreatic neoplasms based on their biological behaviour and prognosis, and ii) to identify imaging markers for early assessment of neoadjuvant treatment response for PDAC and oesophageal cancers. Methods: Five observational studies were designed. Preliminarily, a common radiomic workflow was defined using i) minimum redundancy, ii) robustness against delineation uncertainty and iii) a machine learning bootstrap-based method. #1 – A retrospective set of PDAC patients (pts) who underwent upfront surgery was enrolled to develop a preoperative model to stratify the chance of early (<11 months) distant disease relapse. #2 – A prospective set of IPMN pts candidate for resection was enrolled to identify radiological marker(s) to distinguish low- vs. high-risk IPMN. Besides high-risk stigmata/worrisome features, pancreatic fat content was assessed. #3 – A retrospective set of resected panNEN pts was enrolled to train and validate radiomic models to predict pathological characteristics of aggressiveness. #4 – A retrospective set of PDAC pts who underwent surgery after nCT was enrolled to develop three clusters of models based on a) pre-/ b) post-nCT and c) ∆ radiomics to predict disease recurrence after surgery, N2 and pathological response to nCT. #5 – A prospective set of oesophageal cancer pts scheduled to nCRT was enrolled. PET/MR was performed prior to, during and after treatment; for each patient, ERI was computed. TRG=1 stands for complete response. Results: Models comprising radiomic features and clinicoradiological variables were effective in stratifying the risk of early distant relapse after upfront surgery for PDAC, as well as pathological characteristics of panNEN aggressiveness. ∆ radiomics well stratifies the risk of relapse after nCT for PDAC. Non-radiomic findings were also intriguing. Pancreatic fat content was higher in IPMN pts with high-grade dysplasia/invasive carcinoma than in low/moderate-grade dysplasia. When dealing with cohort#5, pts with TRG=1 had significantly lower ERI values than those with TRG≥2; they also had a significant higher median increase in tumour ADC from baseline to intermediate scans. Conclusions: Our findings could complement current diagnostic workflow and improve clinical decision-making for pts with pancreatic/esophageal neoplasms.
Obiettivi: i) Stratificare in maniera non invasiva le neoplasie del pancreas sulla base di comportamento biologico e prognosi, e ii) identificare marcatori imaging in grado di garantire una valutazione precoce dell’efficacia di trattamenti neoadiuvanti per PDAC e tumori dell’esofago. Metodi: Preliminarmente, abbiamo implementato una workflow comune per l’estrazione di features radiomiche usando i) metodi per garantire robustezza nonostante possibili differenze tra osservatori e ii) un processo tipo machine learning basato sul bootstrap. #1 – Una coorte retrospettiva di pazienti (pz) PDAC sottoposti a chirurgia diretta, arruolata per sviluppare un modello preoperatorio in grado di predire il rischio di ripresa precoce di malattia (<11 mesi). #2 – Una coorte prospettica di pz IPMN candidati a chirurgia, arruolata per identificare marcatori radiologici in grado di distinguere tra IPMN a basso/alto rischio. #3 – Una coorte retrospettiva di pz panNEN resecati, arruolata per costruire e validare modelli radiomici in grado di predire caratteristiche di aggressività biologica. #4 – Una coorte retrospettiva di pz PDAC sottoposta a chirurgia dopo nCT, arruola per sviluppare tre set di modelli basati su radiomica a) pre-/ b) post-nCT e c) ∆ in grado di predire il rischio di ripresa di malattia, uno status linfonodale N2 e la risposta patologica a nCT. #5 – Una coorte prospettica di pz con tumore dell’esofago candidati a nCRT. Ciascun paziente ha eseguito PET/MR prima, durante e dopo nCRT, calcolando l’ERI. TRG=1 significa risposta completa. Risultati: Modelli composti da variabili radiomiche e clinicoradiologiche si sono dimostrati efficaci nello stratificare il rischio di ripresa precoce di malattia per PDAC e nell’identificare caratteristiche di aggressività biologica di panNEN. La radiomica ∆ stratifica bene il rischio di ripresa di malattia (PDAC) dopo nCT. Per quanto riguarda i risultati non radiomici, abbiamo osservato un contenuto adiposo del pancreas maggiore nei pz IPMN con displasia severa/carcinoma invasivo. In riferimento al gruppo #5, pz con TRG=1 avevano valori di ERI significativamente più bassi di quelli con TRG≥2; gli stessi pazienti avevano anche un precoce incremento dell’ADC (alla valutazione intermedia). Conclusioni: I nostri risultati potrebbero integrare l’attuale algoritmo diagnostico e migliorare di conseguenza l’iter terapeutico dei pz con neoplasia di pancreas/esofago.
Identificazione di marcatori imaging per caratterizzare l’aggressività biologica di neoplasie pancreatiche/esofagee e predire l’efficacia di trattamenti neoadiuvanti: un approccio radiomico traslazionale / Diego Palumbo , 2023 Jan 17. 35. ciclo, Anno Accademico 2021/2022.
Identificazione di marcatori imaging per caratterizzare l’aggressività biologica di neoplasie pancreatiche/esofagee e predire l’efficacia di trattamenti neoadiuvanti: un approccio radiomico traslazionale.
PALUMBO, DIEGO
2023-01-17
Abstract
Aims: i) To non-invasively stratify pancreatic neoplasms based on their biological behaviour and prognosis, and ii) to identify imaging markers for early assessment of neoadjuvant treatment response for PDAC and oesophageal cancers. Methods: Five observational studies were designed. Preliminarily, a common radiomic workflow was defined using i) minimum redundancy, ii) robustness against delineation uncertainty and iii) a machine learning bootstrap-based method. #1 – A retrospective set of PDAC patients (pts) who underwent upfront surgery was enrolled to develop a preoperative model to stratify the chance of early (<11 months) distant disease relapse. #2 – A prospective set of IPMN pts candidate for resection was enrolled to identify radiological marker(s) to distinguish low- vs. high-risk IPMN. Besides high-risk stigmata/worrisome features, pancreatic fat content was assessed. #3 – A retrospective set of resected panNEN pts was enrolled to train and validate radiomic models to predict pathological characteristics of aggressiveness. #4 – A retrospective set of PDAC pts who underwent surgery after nCT was enrolled to develop three clusters of models based on a) pre-/ b) post-nCT and c) ∆ radiomics to predict disease recurrence after surgery, N2 and pathological response to nCT. #5 – A prospective set of oesophageal cancer pts scheduled to nCRT was enrolled. PET/MR was performed prior to, during and after treatment; for each patient, ERI was computed. TRG=1 stands for complete response. Results: Models comprising radiomic features and clinicoradiological variables were effective in stratifying the risk of early distant relapse after upfront surgery for PDAC, as well as pathological characteristics of panNEN aggressiveness. ∆ radiomics well stratifies the risk of relapse after nCT for PDAC. Non-radiomic findings were also intriguing. Pancreatic fat content was higher in IPMN pts with high-grade dysplasia/invasive carcinoma than in low/moderate-grade dysplasia. When dealing with cohort#5, pts with TRG=1 had significantly lower ERI values than those with TRG≥2; they also had a significant higher median increase in tumour ADC from baseline to intermediate scans. Conclusions: Our findings could complement current diagnostic workflow and improve clinical decision-making for pts with pancreatic/esophageal neoplasms.File | Dimensione | Formato | |
---|---|---|---|
Diego Palumbo_PhD Thesis def.pdf
Open Access dal 18/01/2024
Descrizione: Diego Palumbo PhD Thesis def
Tipologia:
Tesi di dottorato
Dimensione
5.13 MB
Formato
Adobe PDF
|
5.13 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.