Sunday, May 18, 2025
5/18/2025
Pathologically Interpretable Computational Imaging Predictor for Response to Total Neoadjuvant Treatment in Rectal Cancers - PROJECT SUMMARY: Of the estimated 46,220 patients who will be newly diagnosed with rectal cancer in 2024, patients with locally advanced rectal cancer will undergo total neoadjuvant therapy (TNT) to reduce tumor burden followed by standard-of-care surgical excision of the rectum. Up to 50% of these patients exhibit near complete clinic response (CR, i.e., very few/no tumor cells) on the post-surgical specimen. These patients are therefore ideal candidates for a “watch-and-wait” (W&W) approach, a treatment regimen that replaces unnecessary morbid surgery with intensive surveillance; which results in a significantly improved quality of life while maintaining current disease-free survival rates. However, due to a lack of consistent clinical criteria and variable evaluation of routinely acquired MRIs, the 50% of rectal cancer patients that present with CR cannot be reliably distinguished from non-CR patients. Thus, the key clinical challenge in rectal cancers is accurately and non- invasively identifying patients that exhibit CR after TNT and are candidates for W&W. TNT is known to induce desmoplastic stromal reactions in the tumoral and peritumoral environments due to treatment effects and tumor regression. These pathologic tissue changes are very subtle and difficult to objectively discern on routine MRI. Deep learning (DL), a form of artificial intelligence that utilizes large neural networks to extract textural, morphological, and other attributes, could enable more quantitative characterization of treatment response on imaging. Applying DL approaches for accurate identification of CR in rectal cancers while also maintaining clinical interpretability would require (a) ensuring the DL model uses specialized image filters to capture surrogates of subtle desmoplastic reactions to TNT on routine MRI, (b) biological validation of DL predictions against underlying histology via spatial fusion of MRI and digitized pathology specimens, and (c) rigorous external evaluation of DL performance on a clinical trial cohort of patients who were administered TNT. In this proposal, I will engineer and validate a novel DL framework that leverages wavelet texture filters, called RadWaveNet, to comprehensively quantify TNT response of rectal cancers on MRI. Aim 1 will first develop and optimize RadWaveNet on pre- and post-TNT MRI scans across multiple acquisition planes from a diverse patient population to ensure equity in performance in race- and age-based subgroups. Aim 2a will focus on biologically validating that RadWaveNet signatures captured imaging surrogates of tissue responses associated with CR via spatial co-registration of the post-surgical pathology and pre-operative imaging. In Aim 2b, RadWaveNet will undergo rigorous clinical validation with comparison against gold-standard pathologic response markers for rectal cancer datasets retrospectively curated from the NCT02688712 clinical trial. My project will build upon my promising preliminary results for developing DL models to characterize the rectal lesion environment on MRI, as well as constructing wavelet DL models for predicting response to neoadjuvant chemoradiation; towards a clinically reliable, non-invasive tool for personalizing treatment in rectal cancers.
HHS’ Tracking Accountability in Government Grants System (TAGGS) website provides access to detailed descriptions of grants, loans, aggregated direct payments and other types of financial assistance awarded by the HHS Operating Divisions (OPDIVs) and the Office of the Secretary Staff Divisions (STAFFDIVs).