User:Arturopelayo/Cricothyrotomy Simulator (merged draft)

This is a low-cost procedural-skill trainer for emergency cricothyrotomy, compiled here from the secondary description in D'Auria & Persia (2014)^[1] §2 (the originating laboratory is named under Development Team Contact; the White et al. primary publications were not accessed). Before the attempt the trainee watches an instructional video (New England Journal of Medicine), then on the model palpates the cricothyroid membrane through a bicycle-inner-tube skin, makes a vertical skin incision and a horizontal 1 cm incision in the membrane, inserts a tracheal hook into the cricoid cartilage to open the airway, and places an endotracheal tube. Conductive-foil sensors built into the cardboard trachea and its foam cartilaginous rings record each instrument contact for feedback. The same recorded contact stream can drive an optional automatic-scoring software overlay (D'Auria & Persia's Activity Detection Engine), which checks whether each step is performed correctly and in sequence and gives the trainee immediate feedback (see the "Software Setup" section below). ⚑ Open for review: this merged draft folds the two live pages — White UW (base hardware) and D'Auria (scoring overlay) — into one, because both describe one device from one source (D'Auria & Persia 2014). Whether to merge or keep them split is routed to Catherine; the two live pages stay separate until she rules.

1. Print the cricoid and thyroid cartilages as a single rigid ABS assembly with a mobile cricothyroid joint. STL geometry, print settings, infill, supports, and ABS grade are not recoverable from the accessible secondary source (D'Auria & Persia, 2014)^[1] and remain OPEN pending access to the White et al. primary publications.
2. Fix the printed cartilages to a wooden base so they firmly support the trachea model.

1. Obtain a thin cardboard tube to serve as the tracheal wall, sized to an average adult trachea. The exact inner diameter, wall thickness, and length are not specified in accessible source.
2. Cut compliant-foam strips for the tracheal cartilaginous rings and attach them to the tube with appropriate spacing. Foam grade, ring count, and spacing are not specified in accessible source.
3. Apply conductive-foil strips to the six contact zones labelled A–F per D'Auria & Persia (2014) Figure 2: (A) posterior tracheal wall; (B) right and (D) left lateral trachea and cricothyroid membrane; (C) midline cricothyroid membrane (only correct incision site); (E) cricoid cartilage; (F) cartilaginous ring of the lower tracheal wall.

1. Assemble the ABS cartilage skeleton, cardboard tracheal tube, and foam cartilaginous rings into the cervical-anatomy stack on the wooden base.
2. Drape a segment of bicycle inner tube over the assembled stack to form the skin layer. Inner-tube size, grade, and tensioning method are not specified in accessible source.

1. Connect each of the six conductive-foil contact zones (A–F) to inputs on the Arduino Uno microcontroller board, using a matrix-scanning topology, and wire the procedure instruments (scalpel, tracheal hook, hemostat) into the same circuit so that touching a foil registers as a contact event. Set a 20 ms minimum interval between contacts to debounce the inputs. The detailed wiring layout is not specified in accessible source.
2. The base trainer is now complete and usable for unscored practice (see "Using the trainer" below). To add automatic technique scoring, configure the Activity Detection Engine software overlay as described in the "Software Setup" section.

• Palpate: thyroid cartilage, cricothyroid membrane, and cricoid cartilage identifiable through the bicycle-inner-tube skin — pass/fail
• Sensor continuity: touch each of the six foil zones (A–F) in turn and verify the Arduino registers contact — pass/fail
• Mobile joint: the cricothyroid joint flexes under palpation — pass/fail

The source's intended use procedure (D'Auria & Persia 2014, §2.2):

1. Before the first attempt, watch the New England Journal of Medicine instructional video.
2. Palpate the cricothyroid membrane; immobilise the larynx with the non-dominant hand and perform the procedure with the dominant hand.
3. Incise the skin (bicycle inner tube) vertically after palpating the membrane.
4. Incise the cricothyroid membrane on the trachea model horizontally (1 cm length).
5. Insert the tracheal hook into the cricoid cartilage.
6. Insert the hemostat and expand the airway opening vertically and horizontally.
7. Insert the endotracheal tube.

No Creative Commons licensed figure of this simulator is currently available. Secondary description figures in D'Auria & Persia (2014) are © IEEE All Rights Reserved. Readers with access to the White et al. 2012, 2013, or 2014 primary publications are invited to verify licensing and contribute a compatible image.

Automatic Performance Evaluation — the D'Auria & Persia scoring overlay.

The hardware above is the University of Washington base trainer. The software described in this section is the separate contribution of D'Auria & Persia (2014) — a Cyber-Physical-System (CPS) overlay that adds automatic, real-time technique scoring on top of the same conductive-foil sensors. It is folded in here because both the trainer and the overlay are documented in one source (D'Auria & Persia 2014^[1]); see also the follow-on framework paper.^[2]

While the trainee performs the six-step procedure, the Arduino records each instrument-to-foil contact as a time-stamped event (contact made and contact broken are both logged, in milliseconds). The overlay's Activity Detection Engine takes that time-stamped contact stream and matches it against expert-defined activity models to decide, in real time, whether the technique is being performed correctly and in the right order — giving the trainee immediate correct/incorrect feedback rather than a manual after-the-fact assessment.

The engine compares the contact sequence against Expected Activity models — temporal stochastic automata in which the elapsed time between observations, not just their order, determines whether a sequence counts as a given activity (this is what distinguishes the model from a plain Hidden Markov Chain). Domain experts pre-define the models in two categories: "good" activities (a correct use of the simulator, e.g. an excellent performance) and "bad" activities (an incorrect use). Each model is a labelled directed graph with an initial node, a final node, per-edge upper time bounds, and a probability distribution over each node's outgoing edges. The specific model definitions used in the study are not published in the accessible source.

Activity occurrences in the live contact stream are found using the tMagic algorithm (Albanese, Pugliese & Subrahmanian — indexing for temporal stochastic-automaton-based activity models), which solves the problem of finding occurrences of a high-level activity model in an observed data stream. A Data Acquisition component (with an Adaptation Module) first converts the raw Arduino contact data into a "Simulator Log" in a format the detection framework can consume.

The sensing chain is shared with the base trainer: the Arduino matrix-scans the six foils, a 20 ms minimum interval between contacts debounces the inputs, and the recorded contact stream is low-pass filtered at a 10 Hz cutoff, chosen in line with general human reaction time.

In a reported trial, 100 medical doctors used the simulator with the overlay. Using the standard precision and recall metrics for activity recognition, the engine achieved an average precision of 81% and an average recall of 98%. Separately, participants (classified as expert, medium-expert, and not-expert) completed a 5-point Likert questionnaire on how realistic, anatomically accurate, educational, useful, and real-time they found the simulator; the authors report the satisfaction results as encouraging. These figures measure the scoring engine's recognition accuracy and user satisfaction — not clinical skill transfer to real patients.

1. Confirm the base trainer is built and the six foil zones (A–F) are wired to the Arduino Uno by matrix scanning (Phases 1–3 above).
2. Load the Arduino with firmware implementing the Activity Detection Engine, with a 20 ms debounce interval on each sensor input.
3. Map the 8×8 LED matrix outputs to the six sensor zones and to procedural-state indicators.
4. Provide the Activity Detection Engine with the expert-defined "good" and "bad" Expected Activity models (model definitions are not published in source).
5. Validate the overlay with a dry run of the six-step procedure, confirming that each step triggers the expected LED response and that the engine advances through the procedural states in the correct order.

^{[1][2][3][4][5][6]}