Building The 3-1) Boundary Zones For Predicting Cellular Metabolic Phenotypes By Computation Of Time Sensitivity Factor (S) Using F-18 Fog Pet Potential Preamble Ai Algorithm For Evaluation Of Pulmonary Nodules And Lung Cancer Radiotherapy

1330 Aim: To build 3-D boundary zones for predicting cellular metabolic phenotypes (CMP) of pulmonary nodules (PNs) using F-18 FDG PET by computation of a time sensitivity (S) factor and implement into potential AI algorithm for diagnosis (Dx) and radiotherapy (RT) evaluation of lung cancers (CA). Methods: The initial computation was from 97 PNs (81 patients [pts]) from a US cohort (USC) referred for variable dual-time F-18 FDG PET-CT with definite pathological diagnosis or 1-year CT follow-up (gold standard, GS). S=d{ln[SUV]]/d{ln[T]} was obtained by logarithmic regression using scan times, Ti (i=0, 1, 2), and standard uptake value (SUVj) (j=0, 1, 2) with SUV0 =0.5. The actual T1 and T2 were 72+14 and 135+18 min (Skewness=2.076 and 1.356 respectively) and T0=0.5 min. The subsequent computation was from 41 pts for an Asian cohort (AC) with single PNs imaged by F-18 FDG PET at multiple exact fixed time points (T0=0.5 min; T1-3=1, 2 and 3 hr) with the same GS and technique as in USC for comparison and obtaining the weighted average S into AI algorithm for analysis. The bioinformatics from S computation was finally used to build 3D graph of boundary zones separating the underlying CMP by the zonal plane equation, z= (1+y/x)S , where x= T1, y= T2-x; SUV2 = z[asterisk]SUV1. To help CA Dx, it is necessary to know how long (y) the delayed imaging (DI) needs to be after T1 depending on initial SUV and z by y=x{exp[ln(z)/S]-1}. Without a prior clinical knowledge, the algorithm has to start with a crude PET likelihood of CMP (SUV1>2.5, S at 2 boundary planes) to determine the most useful T2 (x+y) in AI algorithm for PET Dx. Results: There were significant differences in USC between the nonmalignant (NM, n=45) and CA (n=52) groups in S (0.17±0.16 vs. 0.48±0.18, p 2.5 and S=0.45 or SUV1 1.100 (Dx cut off for CA from ROC analysis for AC), then final Dx was reached (group A). Otherwise further DI at 3 hr was needed (Group B). It was determined that if x=60 min(T1) in the AC, 24 out of 41 pts only need T2 at 98.52 min ( 3 hr; so SUV3h, last data point was used for analysis) at the boundary conditions. However, 10 out of initial 24 pts intended for 1.100. The combined group A and B contingency table was generated as: [TP: 26(A=14;B=12)/27], [TN: 13 (A=0;B=13)/14], [FN: 1(A=0;B=1)/27], [FP: 1(A=0;B=1)/14], yielding the overall sensitivity, specificity and accuracy 96, 93 and 95% respectively using this simple AI algorithm. The same algorithm may be useful in evaluation to residual CA cells after RT. As S and its corrected retention index, RIs=[(T2/T1)S- 1] [asterisk] 100%) had been validated previously on micro-PET data during weekly evaluation after hypo-fractionated RT of human lung cancer xenograft to predict tumor control. If one extends this translational research to clinical applications, the initial pre-treatment PET scan may be done at the standard 1- and 2-hr interval at the expected mean RIs = (20.45 - 1) x 100% =37% for CA. After RT, the dual time PET may need to prolong at least or beyond 3 hr to bring out the fundamental changes in CMP from CA cells to fibroblasts and/inflammatory cells due to potentially low SUV and S according to the results of current study with similar AI algorithm. Conclusions: 3-D boundary zones from S computation are useful in AI algorithm for evaluation of CMP of PNs and RT of CA lung.