Enhancing the reliability and accuracy of AI-enabled diagnosis via complementarity-driven deferral to clinicians.

Predictive artificial intelligence (AI) systems based on deep learning have been shown to achieve expert-level identification of diseases in multiple medical imaging settings, but can make errors in cases accurately diagnosed by clinicians and vice versa. We developed Complementarity-Driven Deferral to Clinical Workflow (CoDoC), a system that can learn to decide between the opinion of a predictive AI model and a clinical workflow. CoDoC enhances accuracy relative to clinician-only or AI-only baselines in clinical workflows that screen for breast cancer or tuberculosis (TB). For breast cancer screening, compared to double reading with arbitration in a screening program in the UK, CoDoC reduced false positives by 25% at the same false-negative rate, while achieving a 66% reduction in clinician workload. For TB triaging, compared to standalone AI and clinical workflows, CoDoC achieved a 5-15% reduction in false positives at the same false-negative rate for three of five commercially available predictive AI systems. To facilitate the deployment of CoDoC in novel futuristic clinical settings, we present results showing that CoDoC's performance gains are sustained across several axes of variation (imaging modality, clinical setting and predictive AI system) and discuss the limitations of our evaluation and where further validation would be needed. We provide an open-source implementation to encourage further research and application.

Robust and data-efficient generalization of self-supervised machine learning for diagnostic imaging.

Machine-learning models for medical tasks can match or surpass the performance of clinical experts. However, in settings differing from those of the training dataset, the performance of a model can deteriorate substantially. Here we report a representation-learning strategy for machine-learning models applied to medical-imaging tasks that mitigates such 'out of distribution' performance problem and that improves model robustness and training efficiency. The strategy, which we named REMEDIS (for 'Robust and Efficient Medical Imaging with Self-supervision'), combines large-scale supervised transfer learning on natural images and intermediate contrastive self-supervised learning on medical images and requires minimal task-specific customization. We show the utility of REMEDIS in a range of diagnostic-imaging tasks covering six imaging domains and 15 test datasets, and by simulating three realistic out-of-distribution scenarios. REMEDIS improved in-distribution diagnostic accuracies up to 11.5% with respect to strong supervised baseline models, and in out-of-distribution settings required only 1-33% of the data for retraining to match the performance of supervised models retrained using all available data. REMEDIS may accelerate the development lifecycle of machine-learning models for medical imaging.

Feasibility of IMRT to cover pelvic nodes while escalating the dose to the prostate gland: dosimetric data on 35 consecutive patients.

Utilizing available dosimetric and acute toxicity data, we confirm the feasibility of intensity modulated radiotherapy (IMRT) to include treatment of the pelvic nodes (PN) while escalating the dose to the prostate. Data were obtained from 35 consecutive patients with prostate cancer with > or =15% risk of PN involvement. Patients received an initial boost to the prostate, delivering 16 Gy over 8 fractions using a 6-field conformal technique, followed by an 8-field coplanar inverse planning IMRT technique delivering an additional 60 Gy over 30 fractions to the prostate (76 Gy total) and 54 Gy over 30 fractions to the seminal vesicles (SV) and PN. Dose-volume histogram analysis was performed for planning target volumes and organs at risk. Acute toxicity (RTOG/EORTC scale) was prospectively and independently scored weekly for each patient. The maximum, mean, minimum dose, and D95 to each planning target volume is provided: prostate (82.2, 78.2, 72.6, 75.2 Gy), SV (79.0, 72.5, 56.9, 61.1 Gy), and PN (80.4, 59.7, 46.5, 53.3 Gy), respectively. The percent volume receiving a dose at or above "x" Gy (Vx) was recorded for V75, V70, V65, V60, and V50 as: bladder (14%, 24%, 32%, 39%, and 54%) and rectum (3%, 18%, 26%, 34%, and 51%), respectively. Acute toxicity was as follows: 54% grade 2+ GI (n = 19), 25% grade 2+ GU (n = 9). IMRT enables treatment of pelvic nodes while escalating dose to the prostate and is clinically feasible with acute toxicity within expected ranges.

Are neck nodal volumes drawn on CT slices covered by standard three-field technique?

PURPOSE: Several definitions have been proposed in the past few years on how to contour the various neck nodal levels on CT slices. However, whether the resulting nodal volumes would have been covered by standard techniques is unknown. The purpose of this study was to clarify this issue. METHODS AND MATERIALS: Eight patients (N0-N1) with head-and-neck cancer from various primary sites referred to us for definitive radiotherapy were included in this study. Two observers contoured the level Ib-V neck nodal volumes on planning CT according to seven reported definitions. Each observer also drew blocks on digitally reconstructed radiographs for the initial (on-cord) phase of a standard three-field technique (parallel opposed lateral fields and AP supraclavicular field) for three different clinical settings: "medium" larynx (to cover upper, mid, and low jugular nodes), "big" larynx (same as for medium, plus posterior cervical nodes), and "tonsil" (same as for big plus retropharyngeal nodes). Fields blocks were concentrically reduced 5 mm in all directions as a surrogate for the clinical target volume to planning target volume expansion. A plan was created for each of the clinical settings, delivering 2 Gy to the International Commission on Radiation Units and Measurements reference point. The coverage of the nodal levels according to the various definitions was investigated throughout the analysis of the volume receiving 50%, 80%, and 95% of the prescribed dose (V(50), V(80), and V(95), respectively) and dose covering at least 95% of the volume (D(95)) values extracted from their cumulative dose-volume histograms in the three clinical settings. RESULTS: The V(50) coverage of levels III and IV was adequate for all definitions and trials. For level V, about 3-5% of the volume was outside the 50% isodose of those trials that targeted the posterior cervical chain. Coverage of level Ib was highly dependent on the definition, with up to 21% of the volume outside the standard tonsillar fields. For level II, although the 50% isodose from the tonsillar fields seemed to encompass all definitions, this was not the case for the laryngeal trials. Overall, we found 20-30% of the volumes to be outside the 95% isodose, with larger percentages for levels II and V. Similarly the D(95) results showed all volumes to be underdosed; only about 45% and 65% of levels II and V, on average, received 95% of the prescription dose. CONCLUSION: Within three different clinical settings, we showed that the current definitions provide nodal neck volumes that often fall outside the 50% and 95% isodose lines of the standard three-field technique. Because these volumes are routinely used to define nodal neck volumes for intensity-modulated radiotherapy, the dose-volume objectives of intensity-modulated radiotherapy may not be consistent with those traditionally achieved by the standard three-field technique.

A case of radiation recall mucositis associated with docetaxel.

BACKGROUND: Radiation recall reactions, in particular dermatitis, are well documented in the literature. However, radiation recall mucositis is a rare clinical phenomenon. METHODS: We report a case of a 45-year-old man diagnosed with squamous cell carcinoma of the base of tongue. He was treated with surgery followed by chemotherapy and radiation therapy. Several months after completing treatment, he had a recurrence develop outside of the previously irradiated field. He was offered radiation therapy concurrent with docetaxel as salvage therapy. RESULTS: During salvage therapy, acute recall mucositis developed corresponding to his previously irradiated fields. His chemotherapy with docetaxel was withheld, and his symptoms rapidly improved. CONCLUSIONS: This case describes radiation recall mucositis associated with docetaxel, a rare but potentially serious clinical situation. Given the potential severity of the reaction and increasing use of docetaxel as second-line treatment of recurrent head and neck cancers, it is important to be aware of this phenomenon.

Radiotherapy after prostatectomy.

AIMS AND BACKGROUND: The role of radiotherapy after prostatectomy is controversial. This paper tries to give some guidelines for everyday practice through an analysis of literature data. METHODS: The potential role of radiotherapy in the adjuvant and salvage setting is discussed. We also report and interpret available literature data for both settings. RESULTS: As regards an increase in or detectable prostate-specific antigen (PSA) after radical prostatectomy, about 40-50% of patients are rendered bNED with local salvage radiotherapy, but only 10-50% are long-term (5 years) biochemically controlled. A timely salvage treatment is crucial to optimize control probability. As regards adjuvant radiotherapy for undetectable postoperative PSA in patients at high risk of failure as judged on pathology, results are more encouraging. Recent data report bNED rates > 70% at 5 years. CONCLUSIONS: Although results are far from satisfactory, salvage radiotherapy should be considered for every patient with an increased or detectable PSA after surgery. Adjuvant radiotherapy seems preferable to salvage radiotherapy for patients at high (> 30%) risk of failure.